1

A Noise Power Ratio Measurement Method for

Accurate Estimation of the Error Vector Magnitude
Karl Freiberger, Student Member, IEEE, Harald Enzinger, Student Member, IEEE,
and Christian Vogel, Senior Member, IEEE

Abstract—Error vector magnitude (EVM) and noise power ra- system performance in terms of the bit error rate (BER) [4].
tio (NPR) measurements are well-known approaches to quantify The correlation of EVM and BER can also be understood
the inband performance of communication systems and their as a consequence of two principles of EVM measurements:
respective components. In contrast to NPR, EVM is an important
design specification and is widely adopted by modern com- First, EVM uses the actual communication signal as a test
munication standards such as 802.11 (WLAN). However, EVM signal, exciting the device under test (DUT) in the same
requires full demodulation, whereas NPR excels with simplicity way as during regular operation, which is crucial in case of
requiring only power measurements in different frequency bands. a nonlinear DUT [5]. Second, EVM compares demodulated
Consequently, NPR measurements avoid bias due to insufficient data symbols based on the Euclidean distance of the observed
synchronization and can be readily adapted to different standards
and bandwidths. data symbols from the ideal symbols. Consequently, the EVM
We argue that NPR-inspired measurements can replace EVM is a measure of inband error that allows to use symbol-based
in many practically relevant cases. We show how to set up the correction algorithms. In typical receivers the effect of
signal generation and analysis for power ratio-based estimation linear filtering and common phase error is minimized by
of EVM in OFDM systems impaired by additive noise, power applying equalization and de-rotation, respectively. These two
amplifier (PA) nonlinearity, phase noise and I/Q imbalance.
Our method samples frequency-dependent inband errors via correction steps are also mandatory for EVM measurements
a single measurement and can either include or exclude the according to the 802.11 standard [1].
effect of I/Q mismatch, by using asymmetric or symmetric stop- Although the principle of EVM makes it a sought-after and
band locations, respectively. We present measurement results powerful system level metric, it also comes with the following
using an 802.11ac PA at different back-offs, corroborating the drawbacks. First, EVM is sensitive to synchronization errors,
practicability and accuracy of our method. Using the same
measurement chain, the mean absolute deviation from the EVM e.g., frequency and timing offsets and mismatches [6]. Second,
is less than 0.35 dB. an EVM test setup is expensive [4]. Receiver functionality
is required to test a transmitter, and vice versa. Testing a
Index Terms—NPR, EVM, EPR, nonlinearity, phase noise, in-
termodulation distortion, power amplifiers, wireless LAN, OFDM separate DUT, e.g., an RF PA, even requires a full transceiver.
Advancing to a new standard or bandwidth typically requires
a costly update of the EVM analyzer. Third, it is required
that the whole measurement chain operates linearly at the
I. I NTRODUCTION full signal bandwidth. This gets increasingly difficult with
RTHOGONAL frequency division multiplexing increasing bandwidth [7], limiting the measurement floor at
O (OFDM) is a digital modulation scheme popular in
high-datarate applications like WLAN [1] and LTE [2].
wide bandwidths.
The noise power ratio (NPR) [8]–[11] is a method that does
These standards define maximum error vector magnitude not need demodulated symbols to quantify the inband error. As
(EVM) values for the transmitted signal. The EVM captures a result, the NPR does not suffer from the EVM’s drawbacks
the effect of various transceiver imperfections, e.g., additive outlined above. Consequently, NPR is an appealing alternative
noise, nonlinearity, I/Q mismatch, and phase noise [3]. As a to EVM in many practical scenarios, e.g., as an optimization-
result, the EVM features a strong correlation with the overall objective for digitally enhanced RF systems [12] such as
digital predistorters [13]. However, although the NPR is a
Author version of the manuscript accepted (December 14, 2016) for well established metric on its own, literature does not provide
publication in the IEEE Transactions on Microwave Theory and Techniques.
Date of current version Jan 12, 2017. sufficient support for the following hypothesis: The principle
Karl Freiberger (email: [email protected]), Harald Enzinger, and Chris- behind NPR allows for accurately estimating EVM in OFDM
tian Vogel are with the Signal Processing and Speech Communication Lab systems in case of typical RF transmitter impairments like
(SPSC), Graz University of Technology, 8010 Graz, Austria. Christian Vogel
is with the FH JOANNEUM – University of Applied Sciences Graz, 8020 phase noise, IQ imbalance, and power amplifier nonlinearity.
Graz, Austria. Significant parts of this work were done when the authors were Rather, there is prior research indicating that the NPR is
with the Telecommunications Research Center Vienna (FTW), 1020 Vienna, not able to correctly estimate the inband error caused by a
Austria. This work was supported by the Austrian Research Promotion Agency
FFG (Project Number 4718971). nonlinear DUT [9], [14], discussed in more detail in Sec. II,
c
2017 IEEE. Personal use of this material is permitted. Permission from below. The only work known to us that explicitly addresses
IEEE must be obtained for all other uses, in any current or future media, the relation of EVM and NPR [15] does not cover the issues,
including reprinting/republishing this material for advertising or promotional
purposes, creating new collective works, for resale or redistribution to servers we will show to be relevant for EVM estimation, e.g., the
or lists, or reuse of any copyrighted component of this work in other works. bandwidth and location of the stop bands of the NPR test
 2

signal. The relation of EVM and NPR for IQ imbalance and on the test signal. Rather, there is prior research indicating
phase noise has not been addressed at all so far. quite the opposite [9], [14].
We introduce the term error power ratio (EPR) for our Pedro et al. [14] argue that the NPR underestimates the in-
NPR-inspired EVM estimation method, because we define the band error up to 7 dB, because they view correlated co-channel
EPR as an error to signal ratio, just like the EVM, inversely distortion as a relevant source of error. However, Geens et al.
to the typical NPR definition. Furthermore, we link some [10] reach the conclusion that the approach in [14] is only valid
specifics to the term EPR, e.g., required properties of the test if the input to the nonlinear system varies a lot in amplitude.
signal to obtain accurate EVM estimates for OFDM signals, In most practical cases however, this is not the case and NPR
which further discerns the EPR from the traditional NPR. is a good figure of merit for the inband error [10]. In [17],
In summary, to the best of our knowledge, this paper is the Pedro et al. note that for systems where an equalizer gets rid
first to present: of dynamic linear effects, and hence the correlated co-channel
• The effect of IQ mismatch and phase noise on the NPR. distortion, the NPR is a good choice to assess the inband error.
• An analysis and guidance how to setup NPR (EPR) Gharaibeh [9] shows that the NPR overestimates the inband
measurements to accurately estimate EVM in OFDM sys- error of a (quasi-) memoryless nonlinearity by up to 10 dB
tems with phase noise, IQ mismatch and power amplifier if the excitation signal differs significantly from a circularly-
nonlinearity. symmetric complex normal (CSCN) distribution, where the
• The usage of multiple stop-bands for resolving smooth in-phase (I) and quadrature (Q) component are independent
frequency-dependent errors with a single measurement. and identically distributed (i.i.d) Gaussian [18]. However, the
• A straightforward procedure to obtain an NPR test signal NPR is shown to be a good estimate of inband distortion, if
preserving the statistics of the OFDM signal. the signal approaches a CSCN distribution. In OFDM systems,
• Measurement results comparing NPR (EPR) with EVM, equalizers are ubiquitous and the signal is CSCN distributed.
over a wide range of inband error levels, using a com- Consequently, [9], [10], [14] do not contradict our hypothesis.
mercial hardware WLAN EVM analyzer as reference. Traditional NPR measurements use an analog noise source
The remainder of this paper is organized as follows. Section and an analog filter to generate the test signal [19]. More
II discusses related research. In Section III, we define the recently, digital test signal generation has been proposed [11],
EPR and propose respective test signal generation and analysis allowing for better repeatability and lower test times [8]. A
methods. In Section IV, we present typical transmitter impair- multitone signal is generated via inverse fast Fourier transform
ments and their influence on the EPR and EVM along with (IFFT) of a large number (1k-10k) of tones with random
simulation results supporting our findings. Our experimental phases. Equal magnitudes are used in [11], apart from “5 to
test setup and the analysis of hardware measurement results 10% of the tones” in the center of the bandwidth that are
is presented in Section V. We discuss uncertainty and bias in set to zero to form the stop band. As outlined above, it is
Section VII. Section VII concludes this paper. In Appendix A, desirable to have a test signal matching the statistics of the
we define and discuss OFDM and EVM. communication signal of interest. Therefore, the higher order
cross PSDs of a multitone can be optimized [5], which is
however computationally demanding [20]. For matching the
II. F OUNDATIONS OF NPR M EASUREMENTS AND distribution of each individual sample of an OFDM signal,
R ELATED R ESEARCH such an optimization is not necessary, because summing a
In the following, we discuss prior work on NPR measure- large number of random phase tones yields independent CSCN
ments and EVM estimation. We emphasize why our hypothesis distributed samples. However, OFDM uses a cyclic prefix and
is relevant and that it is neither sufficiently supported, nor oversampling may be applied via zero-padding before the IFFT
contradicted in prior work. [21]. Consequently, the samples are not independent. Rather,
Classic NPR measurements use band-limited Gaussian noise there is correlation, manifesting in a PSD different to the PSD
with a flat power spectral density (PSD) within the band of obtained from a signal with random phases as discussed above.
interest and apply a notch (or band-reject, or stop-band) filter The shape of the PSD is important when assessing the out-
to obtain the test signal [9], [10]. A typical DUT fills the of band behavior, e.g., spectral mask [1] or adjacent channel
stop-band with error, e.g., intermodulation distortion due to a leakage ratio (ACLR) [2]. It is desirable to have an NPR
nonlinear DUT [16]. The NPR is obtained by comparing the signal matching the out-of-band PSD of the OFDM signal,
DUT output power without the filter, or outside the notch, to because this allows for measuring the inband error (NPR) and
the power within the notch, i.e., the NPR is defined as a signal the out-of-band performance with the same test signal. With
(plus error) to error ratio. If the only source of error is additive established NPR test signals [11] this is not advisable, because
white gaussian noise (AWGN), NPR and EVM are inversely they are strictly bandlimited.
related via the signal to noise ratio (SNR). If, however, the EVM estimation has also been addressed in the context of
DUT includes nonlinearity, phase noise or modulator imper- pass/fail tests with automatic test equipment (ATE) [22], [23]
fections, the inband error depends on the test signal exciting with the goal to reduce measurement time and cost. These
the DUT. Since the NPR test signal must include at least one methods differ from our approach by the fact that the type of
inband notch, it cannot be identical to the EVM test signal. DUT is assumed to be known, and an approved prototype of
Therefore, it is not obvious that measurements of NPR can be the DUT is available. The prototype is used to train surrogate
used to replace EVM, in cases when the inband error depends models and decision (pass/fail) rules.
 1

s x y
Source Multi-Stop DUT EPR
Source signal
Source Signal
Ss (f ) Filter EPREPR
testTest
signal
Signal
Sx (f ) DUTDUT
output
Output
Sy (f ) Analysis
0 0 0
EPR

PSD (dB)
−45 −45 −45

M3 , see Fig. 2
−90 −90 −90
−20 10 0 10 20 −20 10 0 10 20 −20 10 0 10 20
Frequency (MHz) Frequency (MHz) Frequency (MHz)

Fig. 1. Principle of the error power ratio (EPR). The EPR is our approach to estimate EVM based on the principle of NPR. The source s is the communication
signal of interest. Depicted is a WLAN signal with 20 MHz bandwidth. We obtain the test signal x by digitally filtering s to attenuate parts of the inband,
achieving steep stop-band notches. The device under test (DUT) introduces errors, which rise the power in the stopbands of the DUT output y. The EPR is
the ratio between average power in the stopbands and average power in the inband of y.

P3 g3
M M3 P4

ωg
a,3 ωa,3 ωc,3 ωb,3 g
ω b,3
0
BWstop,3
−20
PSD (dB)

−40

−60
Sy (ω) g stop,3
−80 BW Sx (ω)

6 6.5 7 7.5 8
Subcarrier Index

Fig. 2. Close-up of the third stopband M3 from Fig. 1. The baseband center frequency is at 2.1875 MHz, i.e., ωc,3 = 7∆f 2π, with the WLAN OFDM
subcarrier spacing being ∆f = 312.5 kHz. We choose the stopband-width for integrating the error power as BWstop = ∆f . The filter stopband M
f3 with
bandwidth BWg stop = 1.3∆f is chosen to be slightly wider to account for the practically limited filter slope steepness.

III. E RROR P OWER R ATIO where λ(K) is the Lebesque measure of the set K, i.e., the
As outlined in the introduction, we use the term error power overall bandwidth of the integration.
ratio (EPR) to distinguish our EVM estimation method from
the traditional NPR [8], [16]. The principle of the EPR is
depicted in Fig. 1. In Fig. 2, we zoom into the third stopband B. Definition
and illustrate the notation used in the definition of the EPR
We define the EPR of a signal x(t) as
below.
S̄x (M)
EPR{x(t)} = , (2)
A. Notation S̄x (P)
Consider a baseband signal x̃(t) modulated to an angu- i.e., we compare the PSD of x(t) averaged over different
lar carrier frequency ω0 = 2πf0 , where t denotes time. frequency sets M and P. Here, M is the set of stop-band
The corresponding radio frequency signal is xRF (t) = frequencies
R{x̃(t) exp ω0 t}, where 2 = −1. The real and imaginary N
[ M

part of x̃(t) are denoted by R{x̃(t)} = x̃I (t), and I{x̃(t)} = M= Mm , (3)
x̃Q (t), respectively. In cases where it is not relevant whether m=1

we are in baseband or RF, we simply use the notation x(t). i.e., the union of NM ∈ N : NM ≥ 1 individual stop-bands
For discrete time signals, we use the notation x[n]. We use Mm . We define the m-th stop-band Mm as the set of angular
boldface notation for vectors e.g., ω = [ω1 , . . . , ωN ]T , where frequencies satisfying
T
denotes transposition.
The power spectral density (PSD) of an ergodic, wide Mm = {ω ∈ R : ωa,m ≤ ω < ωb,m } , (4)
sense
R ∞ stationary random signal x(t) is given as Sx (ω) = where ωa,m = ωc,m −
BW
stop,m
and ωb,m = ωc,m + stop,m BW
2 2
−∞ x
R (τ )e −ωτ
dτ , with the auto correlation function de- are the start and stop frequency of the m-th stop-band with
fined as the expectation Rx (τ ) = E {x∗ (t)x(t + τ )} . We center frequency ωc,m and stop-band bandwidth BWstop,m ,
denote the average PSD of x(t) over a frequency-set K as respectively, as illustrated in Fig. 2. S
Analogously to (4), the set
NP
1 of present bands is denoted as P = p=1 Pp , with Pp = {ω ∈
Z
S̄x (K) = Sx (ω) dω , (1) R : ωa,p ≤ ω < ωb,p }. To compare the EPR of individual
λ(K) ω∈K
 4

stopbands with subcarrier-dependent EVM values as defined i.e., vX = ifft(vXf). However, instead of using constant
in (38) in the Appendix A-B, we define amplitudes and random phases as in [11], we obtain the
frequency-domain vector vXf via an FFT of the communi-
S̄x (Mm )
EPR(km ) = , (5) cation signal vector vX, and set those bins that correspond
S̄x (P)
to stopband frequencies ω ∈ M f to zero before the IFFT,
where Mm has a center frequency at ωc,m = km ∆f , where i.e., vXf = fft(vS); vXf(vIndexNull) = 0. Here,
km is the OFDM subcarrier index belonging to the mth Mf ⊃ M indicates that we set broader bands to zero than the
stopband center, and ∆f is the subcarrier spacing. With equal actual integration range M used to compute the EPR in (2).
stopband bandwidth BWstop,m = ∆f , the overall EPR as Mf includes the same center frequencies ωc,m as M in (4). The
in (2)
Pis the mean over the subcarrier-dependent EPRs, i.e., only difference is that we use BW g stop,m = Kstop BWstop,m ,
1 NM
NM m=1 EPR(km ). Kstop ∈ R : Kstop ≥ 1 to form M, f as shown in Fig. 2.
Our goal is to use the EPR of the DUT response y(t), i.e., If not stated otherwise, we use Kstop = 1.3. This way, we
EPR{y(t)} for estimating the EVM. For that purpose, two accommodate for the limited steepness of the PSD analysis
constraints are necessary in the definition of (2): window, see Sec. III-D, below. Furthermore, using broader
1) Sufficient test-signal stop-band rejection: The EPR of stop-bands increases the robustness against clock frequency
the test signal x(t) used to excite the DUT must be smaller offset.
than the expected error vector power (EVP) to estimate. The The proposed multiplication in the frequency domain with
EVP is the square of the EVM and defined in Appendix zero at the bins to be nulled, and one otherwise, corresponds
A-B. As depicted in Fig. 2 our test signal generation and to a circular (cyclic) convolution in the time domain. This
analysis achieves approximately −90 dB stopband rejection, is welcome in measurement applications, because a typical
which is well below the expected EVM of typical transceiver signal generator repeats the signal vector continuously, i.e., it
components. A stopband rejection of RM = −10 log10 () dB is periodically extended. In contrast to repeating a linearly-
means filtered signal vector, no discontinuities occur due to the
S̄x (M) = S̄x (P) . (6) repetition of the circularly-filtered signal vector. Consequently,
2) Reasonable selection of P and M: To achieve require- we do not require synchronization when analyzing the signal.
ment 1), i.e., small EPR{x(t)}, and to provide maximum In Sec. II, we have already highlighted the importance of
averaging over frequency, P should cover the whole signal using a test signal matching the statistics of the communication
bandwidth with exclusion of the stop-band and transition- signal and that OFDM signals are CSCN distributed. A test
band frequencies M, f as indicated in Fig. 2. As discussed signal xNPR [n] generated with the method in [16] approaches
in Sec. IV, a good choice for the individual stop-bandwidths a CSCN distribution, because the phases are independent
is the OFDM subcarrier-spacing, i.e., BWstop,m = ∆f . The random variables. Since a CSCN signal is invariant to lin-
optimum number and center frequencies of the stop-bands ear filtering [18], also our method delivers a CSCN signal.
Mm is problem-dependent. Setting Nm too high, i.e. using We denote the test signal generated with our method with
to many missing bands increases the likelihood of altering the xEPR [n]. However, even if the distributions of the individual
signal statistics significantly. By setting Nm = 1 frequency- samples of two signals are the same, the PSDs can be quite
dependent errors cannot be resolved in a single measurement different. A signal xNPR [n] generated with i.i.d random phases
and there is only little averaging of missing band power, does not include possible correlations due to cyclic prefix,
which increases the variance of the estimator. A convenient windowed overlap and oversampling via IFFT zero padding.
choice for Nm and ωc is using the pilot-tone locations of Consequently, the PSD Sx,NPR (f ) is strictly band-limited, as
the OFDM signal standard, e.g., subcarriers [−21, −7, 7, 21] shown in Fig. 3. In contrast, the PSD of our EPR test signal
for the 20–MHz WLAN signal in Fig. 1. This way, smooth Sx,EPR (f ) resembles the PSD Ss (f ) of the communication
frequency-dependent errors can be resolved and an increase in signal s(t). Consequently, out-of-band measurements using
bandwidth increases the number of stop-bands. Furthermore, an EPR test signal generated with our approach lead to
the EPR can be readily compared with the EVM averaged only results well comparable with those made with the actual
over the pilots which is provided by typical EVM analyzers. communication signal s(t). Summing up, our signal generation
However, one has to be aware that by using the symmetric approach allows for measurements of out-of-band metrics, e.g.,
pilot locations as stop-band centers, IQ mismatch is excluded spectral mask [1] or ACLR [2], with the same signal that is
from the EPR, as shown in Sec. IV. We further discuss the used for measuring the inband error.
placement of the stopbands in Sec. IV and Sec. VI-B.
D. Signal Analysis
C. Signal Generation To obtain the EPR from the DUT output signal, we propose
We use a digital approach to generate a test signal x[n]. the following two approaches. The first is based on PSD-
In a measurement scenario where an analog signal x(t) is estimation via analyzing the baseband time-domain (TD) DUT
required, we copy the signal vector vX = {x[n]} to the output, whereas the second uses a swept-tuned (ST) spectrum
memory of a signal generator which performs the required analyzer to measure the DUT output power in different bands.
digital to analog conversion (DAC). Similar to Reveyrand The advantage of the TD method is that the same analyzer
et al. , [11], [16], we generate vX by means of an IFFT, code can be used in measurements and simulations. The ST
 5

0
Mask
2) Swept-Tuned (ST) Approach: Swept-tuned analyzers
tune into different analysis frequencies by sweeping the local
−20
oscillator frequency of the down-mixer, from a start frequency
−40 fsw,a to a stop frequency fsw,b during a sweep time Tsw .
This sweep is used to mix the input signal down to an
Sx,EPR (f )
dB

−60
Ss (f ) intermediate frequency (IF). At time t, the signal frequency
−80 mixed to IF, is fsw (t) = fsw,a + fspan mod(t, Tsw )/Tsw ,
where fspan = fsw,b − fsw,a . The center frequency is
−100
Sx,NPR (f ) fsw,c = (fsw,a + fsw,b )/2. After downmixing to IF, the
−5 0 5 10 15 20 signal is filtered with a band-pass filter (BPF) centered at
Frequency (MHz) IF. This BPF determines the resolution bandwidth (RBW)
of the spectral analysis. The power of the BPF output is
Fig. 3. PSDs of three different test signals generated and analyzed on a PC
(not measured). Ss (f ) is the PSD estimate of a 320-symbol WLAN signal
measured using an envelope detector, whose output is further
with 20 MHz bandwidth oversampled in the IFFT to 60 MHz. In contrast smoothed by a so-called video bandwidth (VBW) filter. Swept-
to the traditional NPR signal generation resulting in a strictly band-limited tuned analyzers typically offer several detector modes. To
Sx,NPR (f ), our EPR test signal Sx,EPR [n] features the same out-of-band
behavior as the original communication signal Ss (f ).
minimize bias, an RMS detector should be used not only when
measuring absolute power [26] but also when making power
ratio measurements [27].
method unites all the advantages outlined in Sec. I, when By sweeping the whole bandwidth of the signal with a fine
making measurements. In particular, the ST method does not RBW, it is possible to get a PSD estimate similar to the TD
require analog to digital converters (ADCs) able to sample the method. However, such a measurement can take a significant
entire signal bandwidth with high accuracy. amount of time, e.g., 10 seconds, since the sweep time must
1) Time Domain (TD) Approach: Given be high for high values of fspan and low values of RBW.
T More precisely,
(7) fspan

ỹ = ỹ[n], ỹ[n − 1], . . . , ỹ[n − N + 1] , Tsw = C , (8)
RBW2
i.e., N samples of the digitized baseband-equivalent of the
where C is a dimensionless constant [28]. By tuning into each
DUT output ỹ[n] = ỹ(nTs ), where fs = T1s is the sample rate
stop band separately, we can achieve faster measurements: To
in Hz, we use Welch’s periodogram method [24] to estimate
obtain the average stop-band PSD S̄y (M) in the numerator of
the PSD Sỹ (ωk ) on a discrete frequency grid ωk = k2πf K ,
s
(2), we make NM separate sweeps and set the m-th sweep’s
K ∈ N, k ∈ Z : d−K/2e ≤ k ≤ dK/2e−1, where d·e denotes
start and stop frequency to the m-th stop-band’s start and stop
the ceiling operator. The number of averaged periodograms is ωa,m ωb,m
frequencies, i.e., 2π and 2π , respectively. The RBW must
Navg = b N Kolap c + 1, where K is the FFT length, Kolap is the
−K
be small enough to prevent including power in the slope of
window overlap in samples, and b·c is the floor operator. Once
neighboring present-bands. In our WLAN application we use
we have Sỹ (ωk ) the EPR is computed like in (2), however, due
an RBW of 10 kHz, C = 10, and fspan = 312.5 kHz.
to discrete frequency grid the integrals turn into sums.
To obtain the average present-band PSD S̄y (P), we could
In our WLAN application, we typically use a resolution
proceed similarly as for S̄y (M) above and make NP separate
of fs /K = 10 kHz, a four-term minimum sidelobe Nuttall
measurements. However, we again advocate for an approach
window [25], and 50% overlap, i.e., Kolap = d0.5Ke. This is
decreasing the measurement time: We make a single sweep
also the setting we use in all our PSD plots, e.g., in Fig. 3.
over the whole bandwidth of the signal, since the contribution
Note that if the resolution was too high, e.g., 50 kHz instead
of the missing bands S̄y (M) to S̄y (P) is negligible for typical
of 10 kHz, we would not be able to resolve the steep stopband
inband error values of interest. For instance, if the true inband
notches anymore. For the same reason, a window with good
error is ≤ −20 dB, we obtain −20.04 dB = 10 lg(0.01/(1 +
side-lobe behavior is crucial for the analysis. Given the PSD
0.01)). Since the stopbands do not have to be resolved here,
estimate, we just need to implement (2) by averaging over the fspan
the RBW can be relatively high. We use RBW = 200, yielding
respective PSD frequency bins to obtain the EPR.
RBW = 200 kHz for a signal with a bandwidth of fspan =
As can be seen from Fig. 3, our PSD analysis setup
40 MHz. With C = 100 we still get Tsw in the order of
can resolve EPRs down to −90 dB, which is enough for
100 ms. Though the power is integrated over the full signal
typical transceiver building blocks. With a rectangular analysis
bandwidth fspan , it is important to normalize by the present-
window-length exactly matching the IFFT-length of the signal
bandwidth λ(P) when computing S̄y (P).
generation, there is theoretically no lower limit on the EPR,
i.e., the bins set to zero remain zero if there is no error. Since
the power is integrated in (2), the lack of averaging is not a
problem in practice. However, a slight mismatch of synthesis IV. T RANSCEIVER I MPAIRMENTS AND T HEIR I NFLUENCE
and analysis window-length, e.g., due to mismatches in DAC ON EPR AND EVM
and ADC clock, drastically deteriorates the achievable floor. In
contrast, the proposed windowed PSD estimation is insensitive In the following, we review common analog transceiver
to window-length mismatch. impairments and discuss their influence on the EPR and EVM.
 6

EPR - EVM (dB)

3 B. Nonlinearity
RM = 80 dB
RM = 90 dB In case of nonlinear DUT, e.g., a power amplifier (PA), the
0 distorted DUT output can be decomposed into a correlated
component and an additive uncorrelated distortion noise com-
-3 ponent [9]. The major difference to the additive noise scenario
-90 -80 -70 -60 -50 -40 -30 -20 -10 0 above is that the distortion noise is not independent from
EVM = -SNR = 10log10 (S7v (P)=S7x (P))
the excitation signal x(t), i.e., the excitation signal affects
Fig. 4. Simulated EPR estimation error in case of additive noise for two the effective inband SNR, in general. However, if the EPR
different stopband rejection factors RM . At high SNR (−90 dB EVM), the test signal has similar amplitude statistics like the EVM test
EPR overestimates the EVM, because of the bias due to the limited stopband signal, the effective inband SNR is the same in an EVM
rejection. At low SNR (0 dB EVM) the EPR underestimates the EVM because
we defined the EPR as an error to signal plus error ratio. and EPR measurement, and consequently the EPR is able to
estimate EVM for nonlinear DUTs. To make this more clear,
we investigate a third-order polynomial system
yRF (t) = c1 xRF (t) + c3 x3RF (t) . (14)
A. Additive Noise
Assuming a Gaussian input signal xRF (t), the autocorrelation
function of (14) is given as [10]
Consider the following signal model
Ry (τ ) = (c1 + 3c3 Rx (0))2 Rx (τ ) + 6c23 Rx3 (τ ) . (15)
y(t) = x(t) + v(t) , (9)
Consequently, the PSD of the output signal is
where x(t) = h(t) ∗ s(t), s(t) denotes the signal, h(t) is an
impulse response, and ∗ denotes convolution. Assuming the Sy (ω) =(c21 + 6c1 c3 σx2 + 9c23 σx4 )Sx (ω) + 6c23 Sx,3∗ (ω), (16)
noise v(t) to be uncorrelated with x(t), the PSD of y(t) is with the three-fold convolution term
Sy (ω) = Sx (ω) + Sv (ω) . (10) Sx,3∗ (ω) = (Sx ∗ Sx ∗ Sx )(ω) , (17)
Inserting in the definition of the EPR in (2) yields and the variance of the input signal σx2 = Rx (0) (a zero
S̄y (M) mean distribution of x(t) is presumed for this abbreviation).
EPR{y(t)} = (11) In OFDM receivers, in each frame, new equalizer coefficients
S̄y (P)
are estimated from the preamble. If the source signal and the
S̄x (M) + S̄v (M) nonlinear system are stationary within one frame, which is
= (12)
S̄x (P) + S̄v (P) a reasonable assumption for OFDM and power amplifiers,
Assuming an EPR test signal x(t), (6) holds, i.e., S̄x (M) = the first three terms in (16) cannot be distinguished. Con-
S̄x (P). If we furthermore assume S̄v (M) = S̄v (P) , i.e., the sequently, the whole gain factor c21 + 6c1 c3 σx2 + 9c23 σx4 gets
noise power averaged over the stopband is representative for equalized. Hence, only the uncorrelated distortion 6c23 Sx,3∗ (ω)
the noise at the present bands, (12) becomes contributes to the EVM.
By inserting (16) in the definition of EPR in (2), we see
S̄x (P) + S̄v (P) ξ + 1 1 that the EPR only captures the last term due to the missing
EPR{y(t)} = = ≈ , (13)
S̄x (P) + S̄v (P) ξ+1 ξ excitation at the stop-bands Sx (ω)|ω∈M ≈ 0. This rejection of
where ξ = S̄x (P)/S̄v (P). The approximation in (13) holds correlated distortion is a welcome feature, because it resembles
for   ξ and ξ  1. With SNR = 10 log10 (ξ), RM = the effect of using an equalizer in EVM tests, as discussed
−10 log10 (), and EVM = − SNR as derived in Appendix above. If we have a system with higher order, we get additional
A-C, we obtain the result depicted in Fig. 4. higher-order convolution terms, but the principle remains the
same. Also, the extension to a Wiener-Hammerstein model,
Note that the fixed, systematic bias at high EVMs results i.e., a static nonlinearity between two linear filters introducing
from the definition of EPR as an error to signal plus error memory, is straightforward: Sy (ω) = F {Sx (ω)} in (16)
ratio and could be easily removed, whereas the bias at low becomes
EVM could be decreased by increasing the stopband rejection
Sy (ω) = |Hpost (ω)|2 F |Hpre (ω)|2 Sx (ω) . (18)

RM . In most practically relevant cases, however, covering the
range between −70 and −10 dB EVM is sufficient and bias
Although the input and the output of the nonlinearity are
correction is not necessary.
frequency-weighed, it is clear that (18) can again be decom-
The PSD of the source signal s(t) filtered with h(t) is posed into a correlated term and an uncorrelated term. This
Sx (ω) = |H(ω)|2 Ss (ω). If the filter preserves the power of holds even for more complicated nonlinearites with memory,
the communication signal s(t), the SNR is not affected by h(t) as explained by the Bussgang theorem [9].
and we have S̄x (P) = S̄s (P). Because of the integral nature The EPR is completely blind to the correlated term includ-
of S̄ and the EPR, the exact frequency dependent behavior of ing the correlated distortion, e.g., 6c1 c3 σx2 + 9c23 σx4 in (16)
H(ω) is irrelevant to the EPR. The EPR behaves like data- that arises if we have a non-zero third-order coefficient c3 .
aided EVM with a perfect equalizer. A linear equalizer can compensate the correlated distortion,
 7

−15 −40
−20 EVM EVM −50
−25
dB

without EQ EPR with EQ

dB
−30 −60
−28 −21 −7 0 7 21 28 −70
Frequency (∆f ) −25.5 −25 −24.5 −2 0 2 24.8 25 25.2
0
Sy,EPR (f ) Mask 0
−20 Sy,EVM (f ) EVM without EPR Output
de-rotation PSD Sy (f )
−40
−50
dB

dB
−60 Sy,NPR (f ) Sx,EPR (f ) EPR Input EVM with
−80 PSD Sx (f ) de-rotation
−100
−30 −20 −10 0 10 20 30 −28 −25 −3 0 3 25 28
Frequency (MHz) Frequency (∆f )

Fig. 5. Wiener PA model simulation. Top plot: zoomed version of bottom plot Fig. 6. Phase noise simulation illustrating the importance of the stopband-
with abscissa in multiples of the OFDM tonespacing ∆f . At the stop-band width BWstop for EVM estimation. Bottom: EPR test signal PSD, DUT
center frequencies ωc,M = {−21, −7, 7, 21} · 2π∆f , the EPR resembles output PSD, and EVM per subcarrier. Top: Zoomed missing bands. Left: With
the EVM with equalizer (EQ) at these tones. BWstop = ∆f , i.e., the missing bandwidth equals the OFDM tonespacing,
good estimates of EVM with de-rotation (phase tracking) can be expected.
Middle: With BWstop = 6∆f , the EPR underestimates the EVM. Right:
With BWstop = 0.1∆f , the EPR overestimates the EVM with de-rotation.
since it is, for each OFDM frame, just a weighting factor that
is frequency-dependent for nonlinearities with memory. The
nonlinear nature of the weighting factor is apparent in (16) whereas the equalizer is typically estimated from discrete-time
because it depends on the test signal power σx2 . This is however baseband data. Still, the above argumentation on the EVM and
not a problem for frame-based equalization because σx2 is the equalizer is valid, because when mapping from passband
constant within an OFDM frame. To have an EPR resembling to baseband, the polynomial structure is obtained. Only the
the EVM, the EVM analyzer’s equalizer must be able to coefficient values get scaled with factors depending on their
correct the correlated distortion sufficiently. With OFDM, this order [31], [33]. The autocorrelation for a baseband third-order
is typically the case if the nonlinear memory effects are shorter nonlinearity can be found in (5.37) in [9].
than the cyclic prefix length.
Another important issue with nonlinearities that have mem-
ory is that the error (the uncorrelated distortion) is, in general, C. Phase Noise
frequency-dependent. To estimate EVM correctly, the error Phase noise occurs due to jitter of the local oscillator (LO)
observed in the EPR stopbands must be representative for of the mixer. A baseband model for phase noise is
the error of the whole inband, as we also required for (13).
ỹ(t) = x̃(t)eφ(t) ≈ x̃(t) + x̃(t)φ(t) . (20)
The individual stopband EPRs resemble the EVM at these fre-
quencies which is illustrated in Fig. 5. The overall EPR is the where φ(t) is the phase noise with PSD Sφ (ω). The first-
mean over the individual stopband EPRs as defined in (5). The order Taylor series approximation in (20) is valid if φ(t) <<
overall EVM is the mean over the EVM at all data subcarriers. 1, which is a reasonable assumption for LOs in modern
If the mean over the EVM at the stopband subcarriers differs transceivers [3]. The PSD of ỹ(t) is hence
from the overall EVM, the single-measurement multi-stopband
EPR will also differ from the overall EVM. Sy (ω) ≈ Sx (ω) + (Sx ∗ Sφ ) (ω) . (21)
In practice, most transceiver DUTs feature an error that The additive error term Se,P N (ω) = (Sx ∗ Sφ ) (ω) is statis-
is well-behaved, i.e., not changing abruptly over the inband tically dependent on √Sx (ω). A double-side-band PLL phase
frequencies. If, however, there is reason to expect highly noise model L(f ) = 2Sφ (2πf ) is given as [3]
frequency-dependent error, e.g., when a dynamic element
B 2 L0
 
matching DAC [29] is part of the DUT, there is the possibility fcorner
L(f ) = 2 PLL 2 1 + + Lfloor , (22)
of making several measurements with different stop-band BPLL + f f
frequencies. where f > 0 denotes the frequency offset from the carrier,
Fig. 5 illustrates the EPR for the Wiener PA model from BPLL is the PLL −3 dB bandwidth and L0 is the inband phase
[30]. This model consists of an FIR filter with coefficients noise level in rad2 /Hz, fcorner is the flicker corner frequency,
[0.7692, 0.1538, 0.0769] followed by a quasi-memoryless [31] and Lfloor the noise floor. Depending on the bandwidth of
nonlinear model [32] given as Sφ (ω) compared to the OFDM subcarrier spacing ∆f , two
1.1x̃[n] j0.8 |x̃[n]|
2 cases can be distinguished: Inter-carrier interference (ICI)
ỹ[n] = 2 exp 2. (19) prevails if BWPLL > ∆f /2 and common phase error (CPE) is
1 + 0.3 |x̃[n]| 1 + 3 |x̃[n]| dominant if BWPLL < ∆f /2. While it is very hard to remove
In the simulation belonging to Fig. 5, the EVM without ICI in a receiver, CPE can be mitigated by phase tracking
equalizer is −16.2 dB, whereas it is −32.7 dB with equalizer. [3]. Phase tracking is mandatory for EVM measurements
The EPR is also −32.7 dB, i.e., it equals the EVM with according to the 802.11ac WLAN standard [1].
equalizer. With EPR, the convolution in (21) spills power from
Note that (14) is a continuous-time passband model, neighboring present bands into the stop-band. Since CPE can
 8

out whether IQ imbalance is the limiting factor that determines

0
the inband-error. By contrast, this is not possible with EVM,
−20 since EVM does not allow for excluding IQ mismatch. In
Fig. 7, we illustrate the EVM and EPR for an IQ mismatch
−40 simulation. The used IQ mismatch model is a discrete-time
dB

PSD Sy (f )
Input PSD EVM equivalent of (23) with
−60 after IQ-MM Sx (f ) per subcarrier
h̃I [n] = 0.98δ[n] + 0.02δ[n − 1] (26a)
−80 h̃Q [n] = 0.94δ[n] + 0.06δ[n − 1] (26b)
ϕ
−28 −21 −7 0 7 23 28 ge = 1.01 · exp (π2π/360) (26c)
Frequency (∆f )
where δ[n] is the delta impulse sequence, i.e., we use two-
Fig. 7. IQ mismatch simulation illustrating the importance of the missing band tap FIR filters to model frequency-dependent mismatch of the
locations ωc,m .With the symmetric tones at ±7∆f , IQ mismatch is excluded.
With asymmetric locations (−21, +23)∆f , IQ mismatch is included.
DAC’s I- and Q-path, a gain imbalance factor g = 1.01 and
π degree phase imbalance.
In Fig. 7, we use an EPR test signal Sx (f ) featuring four
be corrected in EVM, we do not want CPE in our stop- stop-bands. Two of these stop-bands (±7∆f ) are symmetric
band, contributing to the error power in the numerator of (2). around zero, whereas the other two are at (−21, +23)∆f , i.e.,
Similarly, we want ICI to contribute to the stop-band power they are asymmetric and do not have an equivalent mirrored
because it is also included in EVM. Since what is considered around zero. The simulation confirms our analysis above:
ICI and what is CPE depends solely on ∆f for a given phase The symmetric stop-bands are blind to IQ mismatch, i.e.,
noise bandwidth, it is clear that also our stop-band width the output PSD resembles the input PSD at these bands.
BWstop must be related to ∆f . As can be seen in Fig. 6, If we only use symmetric bands to compute the EPR, we
using BWstop = ∆f works fine for estimating the EVM with get nearly −90 dB, i.e., the best our analysis window can
activated phase tracking. For Fig. 6, we used rather narrow- achieve, although there is strong IQ mismatch. The asymmetric
band phase-noise with BPLL = 10 kHz, fcorner = 0.5 kHz, bands (−21, +23)∆f , on the other hand, accurately sample the
Lfloor = −150 dB, and L0 = −90 dB in order to clearly frequency-dependent error, depicted as EVM per subcarrier.
see the differences between EVM with and without phase Using only asymmetric stop-bands allows to estimate EVM if
tracking and the influence of different stop-band widths. As IQ mismatch is considered as a part of the DUT. Using both
shown in the measurement chapter V, BWstop = ∆f is symmetric and asymmetric locations at the same time as in
however also appropriate for higher phase noise bandwidths, Fig. 7, allows for checking whether the error is dominated by
e.g., BPLL = 160 kHz. IQ mismatch by just looking at the PSD.
As also discussed in Sec. IV-B, for frequency-dependent
inband error, the mean over the error at the stopbands must
D. IQ Mismatch
approach the overall error. If this is fulfilled the configuration
A baseband model for transmitter IQ mismatch is [34] of the asymmetric stopband is not crucial as long as there is
ỹ(t) = (x̃ ∗ g1 )(t) + (x̃∗ ∗ g2 )(t) , (23) no attenuation of the test signal PSD at mirror-frequencies of
the EPR integration range. In Fig. 7, we used (−21, 23)∆f ,
i.e., x̃(t) = x̃I (t) + x̃Q (t) and its conjugate x̃∗ (t) are i.e. a stopband offset of 2∆f to achieve asymmetry. Using
convolved with impulse responses (IRs) given as only 1∆f would not suffice, because the stopband transition
g̃1 (t) = h̃I (t) + geϕ h̃Q (t) /2 ,


(24a) would create attenuation at a mirror frequency.
ϕ
(24b)


g̃2 (t) = h̃I (t) − ge h̃Q (t) /2 . V. M EASUREMENT R ESULTS
Here, g is the mixer amplitude imbalance factor, ϕ the phase To verify the proposed EPR measurement method experi-
imbalance, and h̃I (t) and h̃Q (t) the IRs of the the in-phase mentally, we made measurements comparing EVM with EPR.
and quadrature path of the D/A converter, respectively. The Our measurement setup is depicted in Fig. 8. We use a PC
power spectral density (PSD) of (23) is connected to the Agilent MXG signal generator and an R&S
2 2
FSQ Analyzer via LAN. The FSQ’s 10 MHz sync output is
Sy (ω) = |G1 (ω)| Sx (ω) + |G2 (ω)| Sx (−ω) , (25) connected to the MXG’s 10 MHz sync input.
where |Gi (ω)| is the Fourier transform of g̃i (t), with i ∈
{1, 2}. If |G2 (ω)| > 0, there is IQ mismatch, and an exci- A. Signal Generation
tation at ω0 causes interference at the mirrored frequency −ω0 . We generate the 802.11ac source signal s̃[n] using a PC run-
With EPR, we can choose where to place the stop-bands. ning MATLAB and the VHT waveform generator tool cited in
Placing them symmetrically around ω = 0, excludes IQ Annex S of [1]. Unless otherwise stated, we use a single burst
mismatch, because then there is no excitation at mirrored with 320 data symbols, 40 MHz bandwidth, and modulation
frequencies. If, however, the stop-bands are chosen to be coding scheme (MCS) 1, i.e., QPSK subcarrier modulation.
asymmetric, the effect of IQ mismatch is included in the EPR Upsampling to a sample rate of 160 MHz is achieved by zero-
result. In practice this can be very helpful when trying to sort padding in the IFFT. For generating EPR excitation signals,
 9

EVMFSQ,pilots , since our EPR stop-bands are also at the pilot

locations, and we expect the FSQ’s phase-tracking algorithm
to achieve the lowest EVM at the pilot-tones.
Our own, data-aided EVM analyzer implemented in
MATLAB analyzes time-domain (TD) baseband data, hence
we refer to it as EVMTD . In contrast to the FSQ’s proprietary
analyzer, we can use our EVM analyzer in simulations, and
have full understanding of the processing. In measurements,
we obtain the baseband data via the FSQ’s I/Q mode data
transfer functionality, i.e., the FSQ downconverts the signal
from 5.6 GHz to baseband and analog to digital conversion.
For EPRTD , i.e., the EPR based on PSD estimation using
Fig. 8. Measurement setup including a PC with MATLAB, Agilent MXG TD data, the measurement chain is exactly the same as
signal generator, and R&S FSQ analyzer. As nonlinear DUT we use an RFMD
RFPA5522 power amplifier (PA) with a 10 dB attenuator at its output.
for EVMTD , only the excitation signal and the analysis is
different. Having the same measurement chain for EPRTD and
EVMTD is beneficial when directly comparing the results.
˜ according to section III-C. We use the standard
we filter s[n]
pilot tone locations {−53, −25, −11, +11, +25, +53} for our D. Calibration
center frequencies ωc,m to obtain symmetric stop-bands. For
The FSQ’s input attenuator (ATT) and vertical reference
asymmetric locations, we offset the positive tones by −2, i.e.,
(VREF) settings are crucial for obtaining best (lowest) and
we use {−53, −25, −11, +9, +23, +51}. After downloading
comparable results. Particularly at EVMs below −30 dB, we
the signal x̃[n] to the MXG, the MXG converts the digital
observed that the FSQ’s automatic ATT and VREF selection
baseband signal to an analog RF signal, centered at 5.6 GHz.
delivers results several dB worse than those obtained with
manually optimized settings. Furthermore, the best settings are
B. Device Under Test not necessarily the same in the different modes (WLAN, TD-
As nonlinear DUT, we use an RFMD RFPA 5522, a IQ, swept-tuned channel power measurement). Accordingly,
commercially available, three stage power amplifier (PA) for each tested power level, we optimized the ATT and VREF
for 802.11a/n/ac applications. To operate the PA in a wide setting to achieve the lowest EVM or EPR, for each method.
range, from a linear regime to deep saturation, we sweep Since the signal path is exactly the same for EPRTD and
through the following 15 MXG analog output gain values: EVMTD , also the optimal ATT and VREF settings are the
{−30, −25, −20, −15, −12.5, −10, −9, . . . , −1} dB. To test same.
the effect of IQ mismatch and phase noise on EVM and EPR, The generator and the analyzer we used receive yearly
we add IQ-mismatch and phase noise to the baseband signal calibration by the manufacturer. Apart from the selection of
in MATLAB, using the models discussed in Sec. IV. The the ATT and VREF setting as described above, user calibration
IQ-mismatch model is given in (26). For the phase noise is not very critical, because we are using a signal analyzer and
profile in (22), we used the parameters BPLL = 160 kHz, not a network analyzer. Furthermore, we are only interested in
fcorner = 2 kHz, Lfloor = −170 dB, and L0 = −93 dB. comparing different methods (EVM, EPR) and not so much in
To get the lowest measurable inband error for reference, we the absolute accuracy, e.g., regarding the power measurement.
made measurements connecting the MXG’s RF output directly The different analysis methods (EVM, EPR) themselves do
to the FSQ’s RF input via an RF cable. However, we were able not need absolute power, but power ratios. Consequently,
to measure the lowest inband error by including the PA and systematic errors in the absolute power level are excluded
the 10 -dB attenuator in the measurement chain and drive it from the result in the ordinate of Fig. 9 by design. The
at a low level (around 2 dBm), because the PA is linear at absolute power measurements varied up to 0.3 dB between the
low levels. The resulting reference (best) RF chain results are different analysis methods. We removed this variation from
summarized in Table I and correspond to the leftmost data our results by using the same data (the power measured with
points of the PA power sweep depicted in Fig. 9. the EVMTD ) for the abscissa of all the different methods in
Fig. 9. This is valid, because of the following: All excitation
C. EVM and EPR Analysis Methods signals are scaled digitally to have exactly the same power.
The measurement chain (generator, cables, PA) is always the
We use the FSQ analyzer in three different measurement
same and differences in output power due to (slightly) different
modes, depending on the EVM or EPR measurement ap-
signal statistics (e.g., PAPR) are negligible. Consequently, it
proach, as described in the following. To have a reference
is reasonable that the power at the output of the DUT is the
to compare our own EVM analyzer code implemented in
same for all analysis methods.
MATLAB, we used the K91ac EVM analyzer firmware
running on the FSQ. We use phase tracking, and make
channel estimation on preamble and data. We refer to the E. Results
resulting average EVM over all data subcarriers as EVMFSQ . The results of the PA power sweep are depicted in Fig. 9.
Furthermore, we are interested in the EVM at the pilot tones The wide range of tested EVM conditions can be seen from
 10

−10 −10
like EVM, asymmetric EPR
−20 −20
Inband Error (dB)

EVMFSQ includes IQ-mismatch

−30 EVMFSQ,pilots −30
EVMTD
−40 −40

−50 −50 symmetric EPR

excludes IQ-mismatch
−60 −60
2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
PA Output Power (dBm) PA Output Power (dBm)
(a) (c)
3 0
2
−0.5
Deviation (dB)

1
0 −1 EPRTD,asymm
EPRST,asymm
−1
−1.5 EPRTD,symm
−2 EPRST,symm
−3 −2
2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
PA Output Power (dBm) PA Output Power (dBm)
(b) (d)

Fig. 9. Inband error (IBE) measurement results comparing several methods/settings for EVM and EPR. The DUT is a WLAN power amplifier (PA). The
signal generator’s output gain is swept. (a) PA measurement. (b) Same data as (a), but deviation from EVMFSQ is shown. (c) PA measurement with IQ
mismatch added to the test signal x[n]. (d) Same data as c), but deviation from EVMFSQ is shown.

Fig. 9 (a). At the highest level (around 30 dBm), the PA is TABLE I

in saturation and the distortion leads to −15 dB of inband RF C HAIN : M EASURED I NBAND E RROR ( D B)
error. At low levels, the PA behaves very linear, achieving
EVMFSQ EPRST EVMTD EPRTD
EVMs down to −53 dB. In essence, the results of all methods
agree over the whole power range. To see differences in more -51.5 -56.3 -47.2 -47.2
detail, the deviation from EVMFSQ is plotted in Fig. 9 (b).
EPRTD and EVMTD have an absolute difference less than TABLE II
0.35 dB over the whole test range. Since the EPR obtained via P HASE N OISE : M EASURED I NBAND E RROR ( D B)
swept-tuned (ST) RF power measurements allows for a lower
M ethod EVM EPR
measurement floor, the deviation defined EPRST,deviation =
EPRST − EVMFSQ is negative for low PA output powers. In FSQ/ST -38.4 -38.9
Fig. 9 (c), we present results of the same PA power sweep TD -38.1 -38.5
as in (a) but with IQ mismatch added to the test signal SIM -38.9 -39.0
x[n]. With symmetric stopbands, the EPR does not capture FSQ, pilots -39.0
IQ mismatch. Therefore, EPRsymm in Fig. 9 (c) agrees with
the result without added IQ mismatch in subplot (a). With
TABLE III
asymmetric rejection bands, IQ mismatch is included and IQ M ISMATCH : M EASURED I NBAND E RROR ( D B)
hence EPRasymm resembles the EVM in Fig. 9 (c).
M ethod EVM EPRasymm EPRsymm
Table I presents inband error results of the RF chain pre- FSQ/ST -35.8 -35.9 -56.3
sented in Sec. V-B, i.e., at low PA output power (2 dBm). The
TD -35.8 -35.6 -46.8
values of EVMTD and EPRTD equal, which indicates that if
the measurement chain is the same, the EPR is able to estimate SIM -35.8 -35.7 -88.6
the EVM very well. With the swept-tuned method EPRST a
lower noise-floor of −56.3 dB can be achieved. In the case of the TD RF signal path. The results in Table III confirm our
of added phase noise shown in Table II, EPRST resembles findings in Sec. IV regarding IQ mismatch. With asymmetric
EVMSIM and EVMFSQ,pilots closely. As expected, the EVM stopbands, however, the EPR results resemble the EVM results
result at the data tones is slightly (0.6 dB) worse. EVMTD and for IQ mismatch. Symmetric stopbands, however, lead to EPR
EPRTD are slightly biased to higher values by the higher floor results resembling the case without added impairment.
 11

VI. D ISCUSSION B. Bias and Stopband Selection

A. Measurement Uncertainty In the following, we discuss bias, i.e., the systematic

deviation of the EPR from the true EVM value. Since, in
In the following, we discuss the uncertainty involved in EPR general, the bias depends on the DUT, it is difficult to
measurements. Uncertainty is the doubt about the measure- remove. However, bias can often be avoided by setting up
ment result [35], i.e., a measure of the variation to expect when the EPR measurement right. Below, we discuss both sides of
making repeated measurements affected by random errors. the following tradeoff. Using only few stopbands is beneficial
Uncertainty analysis of EVM measurements is discussed in in order that the same error occurs in response to the EPR
[7], [36], and references [20-29] in [7]. We discuss systematic test signal as in response to the EVM test signal. However,
errors and bias, i.e., the expected deviation from the true value, for heavily-frequency-dependent error it is important that we
separately in Sec. VI-B. We define three classes depending on observe the error at many different frequency-locations, i.e.,
the source of uncertainty. These are use many stopbands. Our experience is that using multiple
1) Randomness of the excitation signal. stopbands, each one subcarrier spacing broad, with an overall
2) Randomness of the DUT. stopband width of 3 to 10% of the original signal bandwidth is
3) Randomness of the remaining measurement chain. a very good compromise, in practice (six stopbands at 40-MHz
WLAN correspond to about 5%). Choosing the width of each
All three types of uncertainty can affect the repeatability of stopband to be the OFDM subcarrier spacing is important to
the measurement. If we use a fixed test signal, which we do avoid bias in case of phase noise. Placing the stopband-centers
in our measurements, there is no uncertainty of the first type. at OFDM subcarrier-frequencies is convenient for comparison
However, there can be bias, as discussed in Sec. VI-B. with subcarrier-dependent EVM. Then, the individual EPRs
Uncertainty of the second type arises due to limited ob- at each stopband resemble the EVMs at these subcarriers, i.e.,
servation (time and bandwidth) of random DUT errors, e.g., EPR(km ) ≈ EVM(km ). For nonlinear DUTs this is, however,
thermal noise of a PA. The EPR has more type-2 uncertainty only true if the EPR test signal statistics are sufficiently similar
than EVM, because the EPR observes the error only in a to the EVM test signal statics. While for static nonlinearities
fraction αBW = λ(M) BW < 1 of the bandwidth BW used preserving the amplitude statistics (the PDF) is sufficient, for
for averaging the error in EVM. If there is only type-2 nonlinearities with memory it can be important to also preserve
uncertainty, the variance of EPR is hence 1/αBW > 1 times the correlation and higher order moments of the test signal.
the variance of EVM, if the measurement duration is the same. However, retaining the PDF, e.g., a CSCN distribution, does
For deterministic error, e.g., a nonlinear DUT and negligible not mean that the EPR test signal has the same PSD as the
type-1 and type-3 uncertainties, repeated measurements have source signal (and consequently also not the same autocor-
the same result, so both EPR and EVM have zero variance. relation). Quite the contrary, it cannot have the same PSD,
The type-3 uncertainties of EVMTD and EPRTD are the since we need a PSD with stopbands for EPR, whereas the
same, because the measurement chain is the same. Although EVM test signal features no stopbands. The more bandwidth
we do not know the internals of the FSQ and the EVM we null for EPR, the less will the PSDs resemble, and the more
analyzer software, we expect the EVMFSQ to have similar we are risking to produce a different amount of distortion at
uncertainty as EVMTD . For EPRST , a different detector the output compared to the EVM signal which may lead to a
(RMS instead of sample) is used and random errors in the RX biased EPR. However, we observed in many experiments with
IQ path are excluded. Still, our experience is that the difference measured and simulated systems, that this issue is not very
in uncertainty of EPRST compared to EPRTD is small, in critical in practice when nulling, e.g., 10% of the bandwidth.
practice. This is also supported by the informal experiment A related issue is that the test signal should not be too short in
described below. Apart from excluding synchronization errors order to excite the DUT in the same way as a standard EVM
potentially biasing the result, the swept-tuned principle allows test signal. We used 320 symbols according to the EVM test
for EPR measurements with lower measurement floor at high WLAN standard, i.e., 1.28 ms.
bandwidths. Now we discuss the issue of observing the error adequately,
To get more insight for the amount of type-3 uncertainties i.e., the observed error must be representative for the overall
and the repeatability of our measurements, we made ten error. When the frequency-dependence of the error is mild
successive trials of the power sweeps in Fig. 9 with fixed (which is the case for most practical systems) the above
excitation signals. The deviation from the mean (in dB) was presented stop-band placement works well for estimating the
low for all methods, over the whole power range. For the EPR EVM. Then the exact number (between 3 or 8% of the number
methods the deviation was within ±0.17 dB over the whole of number of OFDM subcarriers) and absolute location of the
range. For EVM it was in the same range (±0.12 dB) apart for stopbands is also not critical. We proposed to use to use the
an outlier deviating up to −0.32 dB for higher levels. At lower pilot locations as stopband-centers, because they are easy to
levels, the uncertainty of the EPR measurements was slightly remember and the result can be directly compared to EVM
higher compared to EVM. This is reasonable because at low averaged at the pilots, which is provided by most commercial
levels, the influence of additive noise is stronger and hence EVM analyzers. Sometimes, the error increases or decreases
we have type-2 uncertainty, leading to an increased variance at the band-edges or around DC, so it can be useful to include
of EPR compared to EVM. the outermost and innermost modulated subcarriers. Having a
 12

look on the individual EPR values is always advisable. If the otherwise stated, we assume L = {0, . . . , NL − 1}. The lth
EPR varies a lot, it is sensible to make a second measurement OFDM symbol is given as
with different stopband locations. If the result differs from X  2πk t
the first one, it makes sense to make several measurements, s̃l (t) = w(t) S[l, k]e TK , (28)
sweeping the stopband locations over the whole inband. k∈K

where w(t) is a window function

P with finite support −Tw −
VII. C ONCLUSION Tg ≤ t ≤ TK + Tw fulfilling l∈Z w(t − lT ) = 1 [21]. S[l, k]
We have presented an NPR measurement method with the is a complex {data, pilot, null} symbol modulated on the kth
goal of estimating EVM. To discern our method from the subcarrier of the lth OFDM symbol. The subcarriers (tones)
traditional approach with a single, broad notch in the center of K indexed by k ∈ K = {−NK /2, . . . , NK /2 − 1}, may
the spectrum, and to account for the fact that NPR is inversely be divided into disjoint subsets for data carrying, pilot, and
defined, we introduced the term error power ratio (EPR) for unused (null) tones D, P, and U, respectively. The associated
our NPR-based EVM estimation method. EPR is an attractive cardinalities are ND , NP and NU , respectively, with NK =
alternative to EVM because, in contrast to EVM, EPR does not ND +NP +NU . In the 40-MHz WLAN standard [1], we have
need high-accuracy demodulation and digitization of the whole NK = 128, ND = 108, NP = 6, P = {±53, ±25, ±11},
signal bandwidth. Rather, a standard swept-tuned spectrum U = {−64, ±63, ±62, ±61, ±60 ± 59, ±1, 0}, NU = 14.
analyzer is all that is needed to make high accuracy, low-floor
measurements more or less independent of the bandwidth.
EPR can be a good estimate of EVM (with equalizer B. Error Vector Magnitude (EVM) Definition
and phase de-rotation employed) in case of additive noise,
nonlinearity, phase noise, and IQ mismatch. We have explained The DUT output symbols are given as
this by analytical considerations illustrated with exemplary Z lT +TK
simulation results, and verified by measurements. Besides esti- 1 − 2πk (t−lT )
Y [l, k] = ys (t)e TK dt , (29)
mating EVM by using asymmetric stop-bands, EPR provides TK t=lT
the possibility to exclude IQ mismatch by using symmetric where ys (t) indicates that the DUT output y(t) must be
stopbands. This can be handy when tracking the source of synchronized to s(t). For instance, a potential time-delay must
limited EVM performance or when measuring EVM of DUTs be compensated, e.g., with the method discussed in [37]. In
with very low EVM, without biasing the result with non-ideal practice, an FFT is used to compute Y [l, k] from ys [n]. The
IQ-modulation and/or -demodulation. constellation symbol error is defined
We are convinced that the presented EPR method can be
a valuable tool for RF engineers in the lab when trying to E[l, k] = Yc [l, k] − X[l, k] , (30)
estimate the EVM performance of transceiver building blocks, where Yc [l, k] is a corrected version of Y [l, k]. In data-aided
such as power amplifiers. Although we focused on WLAN EVM analysis, the subcarrier modulation symbols X[l, k]
signals, our method should be readily applicable to other are the known, source symbols, i.e., X[l, k] = S[l, k]. The
OFDM-based communication standards. However, for signals required processing to obtain Yc [l, k] from Y [l, k] corresponds
that are not circularly symmetric normal (CSCN), some further to the signal enhancement facilities of a receiver: To remove
work is necessary. Straightforward linear filtering changes the linear filtering effects, a frequency-dependent equalization
statistics of a non-CSCN signal, in general. Hence, a key issue factor CEQ [k] ∈ C is introduced. Potential degradation due
when trying to estimate EVM for non-CSCN signals is how to to a common phase error (CPE) is combated using a symbol-
generate a signal retaining the communication signal statistics dependent phase de-rotation factor CCPE [l] ∈ C. With that, the
but featuring the required stop-bands. observed constellation symbol enhanced by equalization and
phase tracking is given as
ACKNOWLEDGMENT
Yc [l, k] = CEQ [k] CCPE [l] Y [l, k] . (31)
The authors would like to thank Intel Connected Home
Division (CHD), Villach, Austria, for their guidance, support With data-aided analysis, i.e., the transmitted data symbols
and access to laboratory equipment. X[l, k] are available in the analyzer and are used as a refer-
ence, the equalizer and de-rotation coefficients can be found
A PPENDIX A by minimizing the error in a least squares sense:
OFDM AND EVM X 2
CEQ,opt [k] = argmin |CEQ [k]Y [l, k] − X[l, k]|
A. Orthogonal Frequency Division Multiplexing (OFDM) CEQ [k]
l∈L
A baseband OFDM signal s(t) : R → C can be written as k]Y ∗ [l, k]
P
X[l,
= Pl∈L ∗ (32)
l∈L Y [l, k]Y [l, k]
X
s̃(t) = s̃l (t − lT ) , (27) X
l∈L CCPE,opt [l] = argmin |CCPE [l] Yc,EQ [l, k] − X[l, k]|2
CCPE [l] k∈S
where T = Tw + Tg + TK is the symbol duration consisting
∗
of guard Tg , window Tw and effective symbol time TK [21].
P
X[l, k]Yc,EQ [l, k]
= P k∈S ∗ (33)
The symbol index set L is a connected set of integers. Unless Y
k∈S c,EQ [l, k]Yc,EQ [l, k]
 13

with Yc,EQ = CEQ [k] Y [l, k], meaning that equalization and To have (43) follow from (42), it is assumed that CEQ [k] =
de-rotation is performed sequentially, in contrast to finding 1/H[k], i.e., the equalizer is able to correct the channel. Since
the joint global optimum which would be much more involved. the error contains only the noise, the data-aided EVM in dB
Furthermore, we implicitly assume already compensated time- equals -SNR averaged over the data bins, which is a well
delay and frequency offset, which allows for simple least- known result [38].
squares optimization of the compensation parameters, whereas
R EFERENCES
the joint determination of optimal compensation parameter
values is a non-convex problem [6]. The mean constellation [1] IEEE Std 802.11ac-2013, “Wireless LAN medium access control (MAC)
and physical layer (PHY) specifications–amendment 4: Enhancements
power over a set of subcarriers C with cardinality |C| is denoted for very high throughput for operation in bands below 6 GHz.”
1 X [2] M. Rumney, Ed., LTE and the Evolution to 4G Wireless: Design and
PSC = PS [k] , (34) Measurement Challenges, 2nd ed. John Wiley & Sons, 2013.
|C| [3] L. Smaini, RF Analog Impairments Modeling for Communication Sys-
k∈C
tems Simulation: Application to OFDM-based Transceivers. John Wiley
where the tone-dependent power PS [k] is obtained by averag- & Sons, 2012.
[4] A. Halder and A. Chatterjee, “Low-cost alternate EVM test for wireless
ing over all symbols, i.e., receiver systems,” in IEEE VLSI Test Symposium, May 2005, pp. 255–
1 X 260.
2
PS [k] = |S[l, k]| . (35) [5] J. Pedro and N. Carvalho, “Designing multisine excitations for non-
NL linear model testing,” IEEE Transactions on Microwave Theory and
l∈L
Techniques, vol. 53, no. 1, pp. 45–54, Jan. 2005.
The error vector power (EVP) is defined as the mean error [6] T. Jensen and T. Larsen, “Robust computation of error vector magnitude
power over all data tones D normalized by the respective for wireless standards,” IEEE Transactions on Communications, vol. 61,
no. 2, pp. 648–657, Feb. 2013.
reference constellation power. [7] K. A. Remley, D. F. Williams, P. D. Hale, C. M. Wang, J. Jargon, and
Y. Park, “Millimeter-wave modulated-signal and error-vector-magnitude
PE D measurement with uncertainty,” IEEE Transactions on Microwave The-
EVP = (36)
P ory and Techniques, vol. 63, no. 5, pp. 1710–1720, May 2015.
√XD [8] W. Kester, “Noise power ratio (NPR) - a 65-year old telephone system
EVM = EVP . (37) specification finds new life in modern wireless applications,” Analog
Devices Inc. Tutorial MT-005, 2008.
If every constellation point occurs with the same probability, [9] K. Gharaibeh, Nonlinear Distortion in Wireless Systems - Modeling and
PXD is an estimate of the average constellation power, which Simulation with Matlab. John Wiley & Sons, 2011.
is a constant for a given modulation format. In practice, for [10] A. Geens, Y. Rolain, W. V. Moer, K. Vanhoenacker, and J. Schoukens,
“Discussion on fundamental issues of NPR measurements,” IEEE Trans-
Nl ≥ 100, PXD is more or less identical to the average actions on Instrumentation and Measurement, vol. 52, no. 1, pp. 197–
constellation power. In any case, we use the exact value of 202, Feb. 2003.
PXD for computing the EVM. We see that EVP is defined [11] T. Reveyrand, D. Barataud, J. Lajoinie, M. Campovecchio, J.-M. Nebus,
E. Ngoya, J. Sombrin, and D. Roques, “A novel experimental noise
as a power ratio. The difficulties and flaws of EVM come power ratio characterization method for multicarrier microwave power
into play because for the error power PED , the demodulated amplifiers,” in ARFTG Conference – Spring, June 2000, pp. 1–5.
symbols Yc [l, k] are needed as can be seen from (30). Conse- [12] B. Murmann, C. Vogel, and H. Koeppl, “Digitally enhanced analog
circuits: System aspects,” in IEEE International Symposium on Circuits
quently, accurate synchronization and equalization is required. and Systems, ISCAS, May 2008, pp. 560–563.
To inspect the distribution of the EVM over the subcarriers k, [13] K. Freiberger, M. Wolkerstorfer, H. Enzinger, and C. Vogel, “Digital
we define predistorter identification based on constrained multi-objective optimiza-
p tion of WLAN standard performance metrics,” in IEEE International
EVM[k] = PE [k]/PXD . (38) Symposium on Circuits and Systems (ISCAS), May 2015, pp. 862–865.
[14] J. Pedro and N. D. Carvalho, “Characterizing nonlinear RF circuits for
their in-band signal distortion,” IEEE Transactions on Instrumentation
and Measurement, vol. 51, no. 3, pp. 420–426, June 2002.
C. EVM for Additive Noise [15] J. B. Sombrin, “On the formal identity of EVM and NPR measurement
methods: Conditions for identity of error vector magnitude and noise
power ratio,” in European Microwave Conference (EuMC), Oct. 2011,
Transforming the linear additive noise model in (9) to pp. 337–340.
Frequency domain yields [16] T. Reveyrand, D. Barataud, J.-M. Nebus, A. Mallet, F. Gizard,
L. Lapierre, and J. Sombrin, “Accurate characterization of intermod-
Y (ω) =H(ω)X(ω) + V (ω) . (39) ulation noise in multi carrier wide band power amplifiers based on
a digital synthesis of pseudo noise gaussian stimuli,” Annales Des
Assuming that the impulse respones h(t) has a discrete time Télécommunications, vol. 61, no. 5, pp. 627–644, 2006.
equivalent h[n] with finite support0 < n < Nh − 1, where [17] J. Pedro, N. Carvalho, and P. Lavrador, “Modeling nonlinear behavior
of band-pass memoryless and dynamic systems,” in IEEE MTT-S Inter-
Nh ≤ Ng must be smaller than the effective OFDM guard national Microwave Symposium, June 2003, pp. 2133–2136 vol.3.
interval, the constellation symbol is given as (9) [18] P. J. Schreier and L. L. Scharf, Statistical Signal Processing of Complex-
Valued Data. Cambridge University Press, 2010.
Y [k, l] =H[k]X[k, l] + V [k, l] . (40) [19] H. Rosen and A. Owens, “Power amplifier linearity studies for SSB
transmissions,” IEEE Transactions on Communications Systems, vol. 12,
where H[k] is the NK point DFT of h[n]. Inserting (40) in no. 2, pp. 150–159, June 1964.
[20] N. Carvalho, K. Remley, D. Schreurs, and K. Gard, “Multisine signals
(30) gives the error for wireless system test and design [application notes],” IEEE Microwave
Magazine, vol. 9, no. 3, pp. 122–138, June 2008.
E[k, l] = Y [k, l]CEQ [k] − X[k, l] (41) [21] K. Witrisal, “OFDM air-interface design for multimedia
= CEQ [k]H[k]X[k, l] + V [k, l] − X[k, l] (42) communications,” Ph.D. Thesis, Delft Univ. of Technol-
ogy, 2002. [Online]. Available: http://theses.eurasip.org/theses/339/
= V [k, l] . (43) ofdm-air-interface-design-for-multimedia
 14

[22] A. Nassery, S. Byregowda, S. Ozev, M. Verhelst, and M. Slamani, Karl Freiberger was born in Graz, Austria, in 1984.
“Built-in self-test of transmitter I/Q mismatch and nonlinearity using He received the Dipl.-Ing. degree in electrical engi-
self-mixing envelope detector,” IEEE Transactions on Very Large Scale neering and audio engineering from Graz University
Integration (VLSI) Systems, vol. 23, no. 2, pp. 331–341, Feb. 2015. of Technology, Austria and the University of Music
[23] E. Acar, S. Ozev, G. Srinivasan, and F. Taenzler, “Optimized EVM test- and Performing Arts, Graz, Austria, in 2010. From
ing for IEEE 802.11a/n RF ICs,” in IEEE International Test Conference, 2010 to 2013 he was with bct electronic GesmbH,
ITC, Oct. 2008, pp. 1–10. Salzburg, Austria, working on acoustic echo cancel-
[24] P. Welch, “The use of fast Fourier transform for the estimation of power lation, multichannel speech enhancement, and acous-
spectra: A method based on time averaging over short, modified peri- tic design for communication terminals. From 2013
odograms,” IEEE Transactions on Audio and Electroacoustics, vol. 15, to 2015 he was a researcher at the Telecommuni-
no. 2, pp. 70–73, June 1967. cations Research Center Vienna (FTW), working on
[25] A. Nuttall, “Some windows with very good sidelobe behavior,” IEEE digitally enhanced transceivers. Since 2015 he works as a researcher and
Transactions on Acoustics, Speech, and Signal Processing, vol. 29, no. 1, lecturer at the Signal Processing and Speech Communication Laboratory, Graz
pp. 84–91, Feb. 1981. University of Technology, where he is currently pursuing a Ph.D. degree. His
[26] M. Bertocco and A. Sona, “On the measurement of power via a super- research interests cluster around signal processing, with a current focus on
heterodyne spectrum analyzer,” IEEE Transactions on Instrumentation digital pre-enhancement of WLAN transmitters and related optimization and
and Measurement, vol. 55, no. 5, pp. 1494–1501, Aug. 2006. measurement problems.
[27] M. Tanahashi and K. Yamaguchi, “Uncertainty of out-of-band distortion
measurement with a spectrum analyzer,” IEEE Transactions on Broad-
casting, vol. 61, no. 3, pp. 532–540, Sept. 2015.
[28] Keysight Technologies, “Spectrum analysis basics, application note
150,” 2016. Harald Enzinger was born in Judenburg, Austria,
[29] I. Galton, “Why dynamic-element-matching DACs work,” IEEE Trans- in 1986. He received the Dipl.-Ing. (FH) degree
actions on Circuits and Systems II: Express Briefs, vol. 57, no. 2, pp. in electronic engineering from the University of
69–74, Feb. 2010. Applied Sciences FH Joanneum in 2009 and the
[30] S. Choi, E. R. Jeong, and Y. H. Lee, “Adaptive predistortion with Dipl.-Ing. degree in information and computer en-
direct learning based on piecewise linear approximation of amplifier gineering from Graz University of Technology in
nonlinearity,” IEEE Journal of Selected Topics in Signal Processing, 2012. From 2012 to 2015 he worked as a reseracher
vol. 3, no. 3, pp. 397–404, June 2009. at the Telecommunications Research Center Vienna
[31] H. Enzinger, K. Freiberger, and C. Vogel, “Analysis of even-order terms (FTW). He is currently with the Signal Processing
in memoryless and quasi-memoryless polynomial baseband models,” in and Speech Communication Laboratory at Graz Uni-
IEEE International Symposium on Circuits and Systems (ISCAS), May verity of Technology where he is working towards
2015, pp. 1714–1717. a Ph.D degree. His research interests include behavioral modeling and digital
[32] A. A. M. Saleh, “Frequency-independent and frequency-dependent non- predistortion of radio frequency power amplifiers and digital enhancement of
linear models of TWT amplifiers,” IEEE Transactions on Communica- mixed-signal circuits.
tions, vol. 29, no. 11, pp. 1715–1720, Nov. 1981.
[33] H. Enzinger, K. Freiberger, G. Kubin, and C. Vogel, “Baseband Volterra
filters with even-order terms: Theoretical foundation and practical im-
plications,” in Asilomar Conference on Signals, Systems and Computers,
Nov. 2016. Christian Vogel was born in Graz, Austria, in 1975.
[34] L. Anttila, M. Valkama, and M. Renfors, “Frequency-selective I/Q He received the Dipl.-Ing. degree in telematics, the
mismatch calibration of wideband direct-conversion transmitters,” IEEE Dr. techn. degree in electrical engineering, and the
Transactions on Circuits and Systems II: Express Briefs, vol. 55, no. 4, Venia Docendi in analog and digital signal process-
pp. 359–363, April 2008. ing from Graz University of Technology, Austria
[35] N. Ridler, B. Lee, J. Martens, and K. Wong, “Measurement uncertainty, in 2001, 2005, and 2013, respectively. From 2006
traceability, and the GUM,” IEEE Microwave Magazine, vol. 8, no. 4, to 2007 Dr. Vogel was Assistant Professor at Graz
pp. 44–53, Aug. 2007. University of Technology, from 2008 to 2009 he was
[36] C. Cho, J.-G. Lee, J.-H. Kim, and D.-C. Kim, “Uncertainty analysis in an Erwin Schrödinger postdoctoral research fellow
EVM measurement using a monte carlo simulation,” IEEE Transactions at the Signal and Information Processing Laboratory
on Instrumentation and Measurement, vol. 64, no. 6, pp. 1413–1418, at ETH Zurich, Switzerland, and from 2010 to
June 2015. 2014 he was key researcher at the Telecommunications Research Center
[37] H. Enzinger, K. Freiberger, G. Kubin, and C. Vogel, “A survey of Vienna (FTW). He is currently head of the degree program electronics and
delay and gain correction methods for the indirect learning of digital computer engineering at FH JOANNEUM - University of Applied Sciences,
predistorters,” in IEEE International Conference on Electronics, Circuits Austria and Adjunct Professor at Graz University of Technology, Austria. His
and Systems (ICECS), Monte Carlo, Monaco, Dec. 2016. research interests include the theory and design of digital, analog, and mixed-
[38] H. Mahmoud and H. Arslan, “Error vector magnitude to SNR conversion signal processing systems with emphasis on communications systems and
for nondata-aided receivers,” IEEE Transactions on Wireless Communi- digital enhancement techniques. He worked on all-digital phased-locked loops,
cations, vol. 8, no. 5, pp. 2694–2704, May 2009. transmitter architectures, digital predistortion, vehicular communications, GPS
receivers, and factor graphs, but is most known for his contributions to
the understanding and further development of time-interleaved ADCs. Dr.
Vogel is author and co-author of more than 70 international journal and
conference papers, holds several international patents, and is the co-author
of five papers that have received best paper awards including the 2009 IEEE
Circuits and Systems Society Outstanding Young Author Award and the 2014
IEEE Circuits and Systems Darlington Best Paper Award.