designfeature By Patrick Robbins, Micronetics Inc

THE DRIVE FOR HIGH-RATE, MULTISTAGE-CONVERTER

APPLICATIONS, SUCH AS DIGITAL RADIO, INCREASES
THE NEED FOR DITHERING. TECHNIQUES ARE AVAILABLE
TO HELP YOU ACCOMPLISH YOUR DITHERING GOAL.

Dithering increases
dynamic range
in digital-radio system
pplications such as test equipment and dig-

A ital radio for electronic intelligence are pushing

the limits for both dynamic range and data
rates. This drive is increasing the need for high-rate,
multistage-converter implementations, so these sys-
tems can reach their full design capacity. Designers
use dithering to increase SFDR (spurious-free dy-
namic range) in digital-radio systems that employ
multistage A/D converters (Reference 1). In these
systems, quantization errors in the conversion
process stack up from stage to stage, often resulting
in sizable spurs. Because designers cannot eliminate
these errors, the design goal must be to randomize
them so that the noise floor effectively absorbs them
(Figure 1). (a)
To accomplish dithering, it is necessary to have
sufficient thermal noise present at the input of the
ADC. Although design alternatives exist for ensur-
ing this goal, the challenge is how to determine
whether dithering is even necessary and, if so, to ex-
amine the alternative means of accomplishing it.
COMPONENT AND ARCHITECTURE ARE BOTH KEY
It is important to look at the ADC, as well as the
architecture of the system in general. The first ques-
tion to ask whether the signal contains enough ther-
mal noise. If so, quantization spurs are not issues;
otherwise, they are. Without a previous analysis, it is
difficult to anticipate the magnitude of these quan-
tization spurs. However, once a designer powers up (b)
the system, they will then become glaringly By randomizing the quantization errors, the spurs
obvious. Figure 1 they generate can move into the noise floor,
A quick analysis saves design effort and time. Con- shown here for a 128-kbps FFT without dither (a) and with
sider a typical high-speed ADC that might require dither (b) (courtesy Analog Devices Inc).
a thermal-noise power of
35 to
40 dBm to ran-
domize the quantization spurs. The designer can po- the receiver’s noise figure to the ambient thermal
tentially use an analog or a digital predictive method. noise power and multiplying it by any gain.
The analog method requires predicting the total re- For example, look at a 70-MHz signal that is be-
ceived noise power. A quick analysis involves adding ing undersampled with a receiver bandwidth of 5
www.edn.com May 26, 2005 | edn 71
designfeature Dithering for digital radio

MHz. Compute the ambient thermal

noise power using Boltzmann’s equation:
N1kTB, where k is Boltzmann’s con-
stant, 1.38065010
23J/K; T is the effec-
tive temperature in Kelvin, and B is the
receiver bandwidth. Therefore, N1
( 1 . 3 8 0 6 5 0 1 0
2 3 J / K ) ( 2 9 0 K ) 
(5 MHz)210
14W210
11 mW

117 dBm. If the receiver has a noise fig-
ure of 10 dB and a gain of 30 dB, the to-
tal noise level, N, is
107 dBm10
dB30 dB
67 dB, which is insuffi-
cient for dithering.
The digital method predicts noise
power by examining expected or meas-
ured BER (bit-error rate) and the lowest
expected signal-power amplitude, S. The
modulation scheme can equate the BER
to EB/N0 (bit-energy-to-noise-density ra-
tio) by examing the waterfall curve (Fig-
ure 2 and Reference 2). By knowing the
data rate, use S to determine EB using The waterfall curve indicates the relationship between BER and EB/N0 for
EBS
10 log(data rate) (Reference 3). Figure 2 differing modulation schemes.
First, calculate EB based on the lowest ex-
pected signal power. Then, from a known er, it appears that insufficient noise pow- which, again, counteracts the benefits of
BER, find the corresponding N0 value, er exists, the designer must ensure suffi- dithering. Generally, designers lowpass-
where N0 is the noise power normalized cient noise power by amplifying the sig- filter the noise to avoid this outcome. The
to a 1-Hz bandwidth. To calculate N, you nal and its accompanying noise or adding sampling rate determines the sharpness
must multiply this value by the noise noise to the signal from a noise source. of the filtering, and it is a common prac-
bandwidth. A quick spectrum analysis of For the first method, if the signal has a tice to undersample the signal. This ap-
the modulated signal can be helpful in fairly narrow expected dynamic range, proach results in aliasing, which causes
determining bandwidth. Figure 3 shows then a static gain stage can probably be ef- the digitized signal spectrum to fold over
a QPSK (quadrature-phase-shift-keying) fective. From a design standpoint, this into lower frequencies, further squeezing
signal with a 2-Mbps data rate. method is easy. The noise figure of the the allowable band of the added noise.
The 3-dB bandwidth of the signal can gain stage is the figure of merit in this The greater the bandlimiting, the less the
give a rough approximation to the asso- case. If the noise figure is high, then it bandwidth available for adding noise, re-
ciated noise bandwidth; from the plot, counteracts the benefits of the dithering. sulting in a higher noise spectral density
this bandwidth is 2.6 MHz. Calculating If your design has a wide dynamic to produce sufficient noise power for ef-
N from No in decibel form, use range, then a static gain stage is probably fective dithering.
NN010log(BW), where BW is the unfeasible because the gain, which needs Designers typically specify noise
bandwidth. Note that, even if the noise to be high enough to produce enough sources in units of spectral density, N0,
bandwidth answer is off by a factor of noise power for dithering, might be ac- but also express them in ENR (excess-
two, this error has only a 3-dB effect on ceptable for weak signals but could result noise ratio). You can calculate N0 from
N; therefore, crude estimates are fine for in compression or distortion of large in- ENR by N0ENR
174 dBm/Hz. A typ-
determining whether dithering is neces- coming signals. A power amp with a 1-dB ical noise diode produces a relatively
sary. As an example, use an S value of compression point to handle all cases small ENR, so a design may need a large

60 dBm; a data rate of 2 Mbps; a QPSK, would probably also have a high noise gain to use such a diode as a noise source.
three-quarter-rate, soft-decision modu- figure, as well as being large, inefficient, Amplifying noise is sometimes difficult,
lation; and a BER of 10
6. You then get and costly. A possible approach is to use because it is a broadband signal and also
the following: EB260 dBm
10 log (2 an AGC (automatic-gain-control) cir- because thermal noise has a high peak-
MHz)
123 dBm/Hz; EB/N06 dB cuit, which tailors the effective gain to the to-rms ratio. Without design diligence,
(from the waterfall curve). Therefore, input level. This method can involve a unwanted resonance can crop up. The
N0
123 dBm/Hz
6 dB
129 dBm/ considerable design effort to ensure that initial step is to calculate the needed gain.
Hz; N
129 dBm/Hz10log (2.6 MHz); the control loop is fast enough and to Using the example of a 70-MHz, un-
and N
65 dBm, which is also insuffi- keep the overall noise figure in check. dersampled signal, the resulting maxi-
cient noise to require dithering. The second method—adding noise to mum noise bandwidth that would be
If this quick analysis predicts that, un- the signal from a noise source—is well- available without encroaching on the sig-
der all expected operating conditions, known but has caveats.You can easily add nal might be about 2 MHz. Using a noise
enough noise power enters the ADC, then noise that is cospectral with the signal, but diode that produces
144 dBm/Hz of
extra circuitry is unnecessary. If, howev- this technique increases the noise floor, noise output (30-dB ENR) and a desired
72 edn | May 26, 2005 www.edn.com
 designfeature Dithering for digital radio

The spectrum of a 2-Mbps QPSK signal shows bandwidth and shape

Figure 3 factors.

integrated noise power for dithering of add to the unit cost in production? How

40 dBm, NN0 10log(2 MHz)G; risky is the design of the dithering cir-
G
40 dBmⳮ(
144 dBm/Hz)
63 cuitry? What are the economic benefits
dB; and G41 dB, where G is gain. The of the program as a result of the increase
next problem involves combining the in dynamic range that the dithering cir-
noise and the signal. To minimize signal- cuit yields? Whether it’s a radio network
path loss, also a figure of merit, a designer requiring fewer base stations as a result
could use a 20-dB coupler, but this ap- of increased range or a piece of test
proach reduces the noise power by 20 dB. equipment gaining a competitive advan-
Thus, the required gain is actually 61 dB. tage, dithering will become increasingly
Note also that some extra lowpass filter- commonplace as its implementation be-
ing may be necessary, again increasing comes simple.왏
the required gain.
Both methods require a large upfront References
design effort and potentially add expen- 1.“Overcoming Converter Nonlinear-
sive circuitry, which leads to the eco- ities with Dither,” Analog Devices Appli-
nomic question: Is it worth it? An alter- cation Note AN-410.
native is to use a self-contained dithering 2. “SHF Interoperability Phase Shift
module, which can tip the scales toward Keying (PSK) and Performance Stan-
employing dithering using the added- dards for SHF Satellite Communications
noise method. This type of device has a PSK Modems,” Military Standard 188-
signal-input port and an output of signal 165, Jan 13, 1995.
plus dithering noise. The noise is suffi- 3. Pearce, Jim, “What’s All This Eb/N0
ciently strong enough and filtered to not Stuff Anyway?” Spread Spectrum Scene
encroach on the signal, even in the pres- Online, Volume 7, No. 1, Fall 2000,
ence of aliasing. The signal-insertion loss www.sss-mag.com/newiss4.html#5.
is less than 1 dB, making it comparable
to a good low-noise amplifier. Having no Author’s bio graphy
active components in the signal path Patrick Robbins is director of the Noise
means that compression or distortion is Products Division of Micronetics Inc
not a problem. (Hudson, NH). He previously worked at
There are always design alternatives to KnowledgeXtensions, developing applica-
any problem. Looking at the big picture, tions for its Java/XML-based e-learning
a quick economic analysis is useful. How software platform. He has a bachelor’s de-
much will the upfront design effort cost? gree in interdisciplinary science from Rens-
How much does the dithering circuitry selaer Polytechnic Institute (Troy, NY).
74 edn | May 26, 2005 www.edn.com