This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2019.2957641, IEEE Internet of
Things Journal
1

NB-IoT Uplink Receiver Design and

Performance Study
Arvind Chakrapani, Qualcomm Flarion Technologies, New Jersey, USA.
Email: [email protected]
Abstract— LTE Narrowband Internet of Things (NB-IoT) is a frequency-division multiple access (SC-FDMA) similar to
3GPP1 defined cellular technology that is designed to enable LTE, with some important changes (see [3]). The NB-IoT
connectivity to many low-cost and low power/throughput IoT uplink has three physical channels defined with a completely
devices running delay-tolerant applications. NB-IoT can coexist redesigned random access (NPRACH) along with data
within LTE spectrum either in a standalone mode, in-band with (NPUSCH Format 1) and control (NPUSCH Format 2)
LTE or in guard-band of LTE. With NB-IoT designed to operate
in very low signal to noise power ratios, the uplink receiver design
channels. The NB-IoT uplink channels can be configured to
presents several challenges. In this paper the design and use different subcarrier spacings4, single or multiple tones and
performance of a NB-IoT uplink receiver is studied in detail. First, repetitions for providing operating flexibility. With NB-IoT
receiver design for NB-IoT uplink channels, with corresponding designed to cater for a maximum coupling loss (i.e. enhanced
mathematical analysis, is presented. Specifically, it is shown how coverage) of up to 164dB in 3GPP release 14 [2], the operating
the time/frequency structure of signals can be exploited to enhance signal-to-noise ratio (SNR) regimes are such that signal power
the receiver performance. Second, the performance of each is well below noise power requiring receiver to boost the
channel is characterized with both link level simulations and received SNR through time/frequency combining. In this paper,
implementation on a commercially deployed Qualcomm® FSMTM the focus is on the design and performance of an NB-IoT uplink
platform2 [1]. Comparisons against the 3GPP defined Radio
receiver.
Performance and Protocol aspect requirements3 are also
Although, there have been several contributions in 3GPP that
provided. Finally, implementation details are addressed and
discussions on proposed enhancements to NB-IoT in 3GPP Release reported the receiver performance of the NB-IoT uplink
15 are provided. It is shown how the proposed receiver algorithms channels, receiver design details have not been provided (see
can be adopted to Release 15 enhancements with minor or no [2] and the references therein). As far as we are aware, the
modifications. The work in this paper is of significance to system receiver design details for NB-IoT uplink shared data and
designers looking to implement efficient NB-IoT uplink receiver to control channels (i.e., NPUSCH Formats 1 and 2 respectively)
coexist with legacy LTE systems. have not been investigated in the current literature. However,
Index Terms—3GPP, NB-IoT, LTE. recently there has been a few studies of the NB-IoT random
access channel (NPRACH) receiver [4]-[7]. An NPRACH
1. INTRODUCTION receiver was proposed in [4], where a joint estimation of
Internet of Things (IoT) will connect numerous heterogeneous preamble detection and timing delay estimation was proposed
devices to the internet; enabling data exchange, processing, using a two-dimensional FFT. However, no discussion is
monitoring, control, etc. resulting in a myriad of use cases. provided on how to derive the threshold for correct detection
Cellular technologies have been developed or being enhanced and implementation issues are not addressed. Moreover, the
to play an indispensable role in the IoT world by catering to complexity of doing a 2-dimensional FFT in a practical system
categories of devices with different capabilities in power, cost is prohibitive. A joint optimization technique was investigated
and complexity. Narrowband IoT (NB-IoT), developed (and in [6] to configure the NPRACH parameters in order to reduce
being enhanced) by 3GPP, promises to provide improved the collision probability under a target delay constraint.
coverage for massive number of low-throughput low-cost However, the complexity of the proposed scheme may not be
devices having low power consumption requirements [2]-[5] feasible for practical implementation, particularly when NB-
with relaxed latency constraints. Prospective applications IoT is required to co-exist with LTE and multiple users of both
include utility metering, environment monitoring, asset LTE and NB-IoT have to be served. In [8], a low-complexity
tracking, municipal light and waste management, etc. The NB- NPRACH receiver was proposed. The proposed method incurs
IoT downlink (DL) and uplink (UL) channels are restricted to a 4 × 𝑅 delay in detection and timing estimation, where 𝑅 is the
180kHz system bandwidth [7], occupying only one physical configured repetitions of the preambles. The delay in preamble
resource block (PRB) of LTE and supported only for frequency- detection could become very large when the number of
division multiplexing (FDD) mode in 3GPP releases 13 and 14. configured repetitions is large. Moreover, the storage
NB-IoT can operate in in-band or guard-band of LTE or in a requirements scale with repetitions configured (see [8] for more
standalone manner. The NB-IoT DL uses the same numerology details). In a real-time system with processing and memory
as the 3GPP Long Term Evolution (LTE) and is based on the constraints, a simpler receiver algorithm is desirable. In this
LTE design of orthogonal frequency division multiplexing work efficient receiver algorithms for all three NB-IoT uplink
(OFDM). The NB-IoT UL is based on a single-carrier channels are proposed with the following contributions. First, a

1
NB-IoT was finalized as part of 3rd Generation Partnership Project (3GPP) 3
Radio Performance and Protocol aspect requirements are specified by the
release 14 specification[2]. 3GPP Radio Access Network working group 4, also known as RAN4.
2 4
Qualcomm FSM is a product of Qualcomm Technologies, Inc. and/or its For NPRACH, only 3.75kHz subcarrier spacing is allowed in 3GPP release
subsidiaries [1]. 14 and can be configured to use either 1.25kHz or 3.75kHz in release 15.

2327-4662 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2019.2957641, IEEE Internet of
Things Journal
2

detailed mathematical analysis is provided with receiver design The NPRACH channel uses only 3.75KHz sub-carrier spacing
guidelines for each NB-IoT uplink channel. It is shown how the and has a group of symbols (denoted as symbol group) each
similarity between NPUSCH single tone data channel and consisting of 5 symbols with a cyclic prefix (CP). Each
NPUSCH control channel can be exploited to use common NPRACH opportunity within an NPRACH resource is confined
estimation modules to reduce memory and implementation to 12 sub-carriers, which can span up to 128 repetitions. To
efforts. Second, discussions are provided on implementation reduce the relative CP overhead, 𝑁 sample OFDM symbol is
issues to be considered in a practical system. Third, link level repeated 5 times and a single CP of length 𝑁𝑐𝑝 is added to form
simulation results are presented along with performance study a symbol group. Two NPRACH Formats, i.e., Format 0 and
on a Qualcomm FSM Chipset based LTE small cell base-station Format 1 are defined for NB-IoT with 𝑁𝑐𝑝 of 66.7 μs and
[1]. The rest of the document is arranged as follows. In section 266.7 μs respectively. In this work the only NPRACH with
2, the NPRACH channel is introduced and a receiver solution preamble Format 0 is investigated, as shorter coverage areas are
proposed. In section 3, the uplink shared channel for data i.e., typically associated with small cells. To facilitate uplink timing
NPUSCH Format 1 receiver is analyzed followed by an analysis estimation at eNB, the single tone preamble hops across
and design of the NPUSCH Format 2 receiver in section 4, different subcarriers. The hopping pattern consists of inner
where most of the estimation modules of NPUSCH Format 1 hopping fixed and outer pseudo-random hopping across
are reused. In section 5, the performance of each of the NB-IoT repetitions (for any repetitions 𝑁𝑅𝑒𝑝 𝑁𝑃𝑅𝐴𝐶𝐻
configured). Inner
channels is provided using detailed link-level simulations hopping is applied between the groups of 4 symbols groups
followed by discussions on implementation complexity and with symbol groups 0,1 and 2,3 being one subcarrier apart; and
practical issues. Comparison of the simulation results with symbol groups 1,2 being 6 subcarriers apart as shown in Figure
RAN4 recommended minimum performance requirements is 2-2. Outer hopping is across repetitions with a different start
provided along with NB-IoT uplink receiver performance on a
subcarrier index chosen from a pre-defined pseudo-random
Qualcomm FSM Chipset based LTE small cell base-station.
sequence (see [11]).
Finally, in section 6, discussions on NB-IoT enhancements
being proposed in 3GPP release 15 are provided and how the
mod  nf , N NPRACH / 10  = 0
proposed receiver can be adopted for the enhancements are  period 
shown.
N NPRACH
period
2. NPRACH RECEIVER DESIGN
Transmitting a random-access preamble is the first step of
NPRACH Radio-
random-access procedure that enables a user equipment (UE) to N start
establish a connection with the network. Apart from accurate frame
preamble detection, estimating uplink timing is another main 47
objective of NPRACH receiver. The acquired uplink timing is
used to command the UE to perform timing advance to achieve
Tones

uplink orthogonalization in OFDMA/SCFDMA systems [3]. NPRACH R1

Nsc
The NB-IoT physical random-access channel (NPRACH)
0
refers to the time frequency resource on which random access
N NPRACH NPRACH
Nrep
preambles are transmitted. The required NPRACH parameters scoffset
configured by higher layers [13] are illustrated in Figure 2-1
and defined as,
𝑁𝑃𝑅𝐴𝐶𝐻 Figure 2-1: NPRACH time/frequency structure
1. 𝑁𝑝𝑒𝑟𝑖𝑜𝑑 (3 bits, nprach-Periodicity) ∈ {40, 80, 160,
240, 320, 640, 1280, 2560} in milliseconds
corresponds to the NPRACH period over which UEs NPRACH (Access)
can do random access.
𝑁𝑃𝑅𝐴𝐶𝐻
2. 𝑁𝑟𝑒𝑝 (3 bits, nprach-NumRepetitions) ∈ {1, 2, 4,
12 to 48 tones

R2 R4
8, 16, 32, 64, 128}, corresponds to the NPRACH pseudo-
preamble repetitions. R3 random
𝑁𝑃𝑅𝐴𝐶𝐻 hopping
3. 𝑁𝑠𝑐𝑜𝑓𝑓𝑠𝑒𝑡 (3 bits, nprach-SubcarrierOffset) ∈ {0, 12, R1
24, 36, 2, 18, 34} corresponds to the subcarrier offset
within the 180kHz bandwidth.
𝑁𝑃𝑅𝐴𝐶𝐻
4. 𝑁𝑠𝑐 (2 bits, nprach-NumSubcarriers) ∈ {12, 24,
36, 48} corresponds to the number of subcarriers being 1 2 3 4 5
used for random access. 1 2 3 4 5
𝑁𝑃𝑅𝐴𝐶𝐻 CP is 266.67 ms or
5. 𝑁𝑠𝑡𝑎𝑟𝑡 (3 bits, nprach-StartTime) ∈ {8, 16, 32, 64, symbol group
6 tones

66.67 ms 1.4/1.6 ms
128, 256, 512, 1024} in millisecond. This corresponds
to the time of the start of the NPRACH transmission is 1 2 3 4 5
𝑁𝑃𝑅𝐴𝐶𝐻
𝑁𝑠𝑡𝑎𝑟𝑡 subframes after the first subframe in radio 1 2 3 4 5
𝑁𝑃𝑅𝐴𝐶𝐻 one NPRACH repetition
𝑁𝑝𝑒𝑟𝑖𝑜𝑑
frames fulfilling 𝑚𝑜𝑑 (𝑛𝑓 , ) = 0. Figure 2-2:NPRACH inter and intra repetition hopping.
10

2327-4662 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2019.2957641, IEEE Internet of
Things Journal
3

NB-IoT has the same OFDM symbol boundary as legacy LTE. A. Frequency offset estimation and correction
That is, in a given subframe, symbol 0,7 have 128 data samples First, 𝜏 is eliminated in equation (6) and the frequency offset 𝜉
with 10 CP samples, and all the other symbols (i.e., 1 to 6 and is estimated by exploiting symbol combining across groups and
8 to 14) have 128 data samples with 9 CP samples with 𝑓𝑠 = repetitions (if configured). Towards this end, the following
1.92 MHz. There is no sampling rate below 1.92 MHz that can steps are performed. Define the product of symbol group 𝑚
always align with the OFDM symbol boundary, i.e., have with 𝑚 ∈ {0,1,2, 3}) and conjugate of symbol group 𝑚 + 1 as,
integer number of samples for both CP and data. This means
that if lower input sampling rate is used, then time-domain 𝑍𝑚 = 𝑌𝑚 𝑐𝑜𝑛𝑗(Y𝑚𝑜𝑑(𝑚+1,4) ) ∝ 𝑒 −𝑗2𝜋𝜉𝑁𝑔 𝑒 𝑗2𝜋𝐻𝑚 𝜏⁄𝑁 (7)
interpolation is needed to adjust for the different sampling time
offset. Sampling at 240kHz will require decimation by 4 with Here, the hopping pattern 𝐻𝑚 is defined as,
two finite impulse response filters. The complexity of 𝐻𝑚 = 𝑛𝑠𝑐 𝑅𝐴 (𝑚 𝑅𝐴 (𝑚)
+ 1) − 𝑛𝑠𝑐
processing samples at 1.92MHz or at 240kHz are very similar with 𝐻3 = 𝑛𝑠𝑐 𝑅𝐴 (0) 𝑅𝐴 (3)
− 𝑛𝑠𝑐
and the details are outside the scope of this paper. In this paper Next combining across symbol groups within inner hopping
the sampling rate for PRACH is assumed to be 𝑓𝑠 = 1.92 MHz. (with 𝑚 ∈ {0,1}) (see [9]), we get,
The derivation below is based on [5] and [9]. Let,
𝑅𝐴 𝑋𝑚1 = 𝑍𝑚 + 𝑍𝑚+2 ∝ 𝑒 −𝑗2𝜋𝜉𝑁𝑔 𝑐𝑜𝑠(2𝜋𝜏 ⁄𝑁).
𝑠𝑚,𝑖 [𝑛] = 𝑒 𝑗2𝜋𝑛𝑠𝑐 (𝑚)𝑛⁄𝑁 , for 𝑛 Combining across repetitions (or over outer hopping) and
(1)
= 𝑁𝑚,𝑖 − 𝑁𝑐𝑝 , … , 𝑁𝑚,𝑖 + 𝑁 − 1 number of receive antennas gives,
𝑅𝐴 (𝑚)
be the transmit symbol at 𝑛𝑠𝑐 subcarrier of 𝑚𝑡ℎ symbol
𝑊1 = ∑ 𝑋𝑚1 ∝ 𝑒 −𝑗2𝜋𝜉𝑁𝑔
group with 𝑁𝑚,𝑖 = 𝑚𝑁𝑔 + 𝑖𝑁, where, 𝑠𝑚,𝑖 [𝑛] – is the time (8)
𝑚∈{0,4,…4(𝑅−1)}
domain waveform at 𝑛𝑡ℎ sample of 𝑖 𝑡ℎ symbol in 𝑚𝑡ℎ symbol 𝑁𝑟𝑥
group; 𝑁𝑐𝑝 is size of CP (16 samples at 𝑓𝑠 = 240𝑘𝑠𝑝𝑠); 𝑁 is where 𝑅 is the NPRACH repetitions configured by upper
size of a symbol (64 samples at 𝑓𝑠 = 240𝑘𝑠𝑝𝑠); 𝑁𝑔 is size of a layers. In the above equation, summation is also over number
𝑃𝑅𝐴𝐶𝐻
group (𝑁𝑐𝑝 + 5𝑁 = 336 samples at 𝑓𝑠 = 240𝑘𝑠𝑝𝑠 = 1.4ms). of receive antennas 𝑁𝑟𝑥 , and 𝑅 = 𝑁𝑅𝑒𝑝 . Similarly, we can
The received signal can be written as, obtain,
𝑍 − 𝑍𝑚+2 𝐻𝑚 = +1
𝑦𝑚,𝑖 [𝑛] = ℎ𝑚,𝑖 𝑒 𝑗2𝜋𝜉(𝑛−𝜏) 𝑠𝑚,𝑖 [𝑛 − 𝜏] 𝑋𝑚2 = { 𝑚
𝑅𝐴 (2) 𝑍𝑚+2 − 𝑍𝑚 𝐻𝑚 = −1
= ℎ𝑚,𝑖 𝑒 𝑗2𝜋𝜉(𝑛−𝜏) 𝑒 𝑗2𝜋𝑛𝑠𝑐 (𝑚)(𝑛−𝜏)⁄𝑁 ∝ 𝑒 −𝑗2𝜋𝜉𝑁𝑔 𝑗𝑠𝑖𝑛(2𝜋𝜏 ⁄𝑁)
where 𝜉 is the frequency offset normalized with sampling
frequency; 𝜏 is the round-trip delay (RTD) normalized with 𝑊2 = ∑ 𝑋𝑚2
𝑚∈{0,4,…4(𝑅−1)} (9)
symbol duration; ℎ𝑚,𝑖 is the channel coefficient at the 𝑖 𝑡ℎ time 𝑁𝑟𝑥
sample of the 𝑚𝑡ℎ symbol group and 𝑗 = √−1. Dropping CP ∝ 𝑒 −𝑗2𝜋𝜉𝑁𝑔 𝑗𝑠𝑖𝑛(2𝜋𝜏 ⁄𝑁)
and taking FFT gives, Note that computing 𝑊2 may not be necessary if the frequency
offset is not anticipated to be high. For example, with
𝑁𝑚,𝑖 +𝑁−1
𝜉 =200Hz, the angular rotation over a symbol group of size
𝑌𝑚,𝑖 [𝑘] = ∑ 𝑦𝑚,𝑖 [𝑛]𝑒 −𝑗2𝜋𝑛𝑘 ⁄𝑁 . (3) 𝑁𝑔 is ≈ 101 degrees. For smaller values of 𝜉, computing only
𝑛=𝑁𝑚,𝑖 𝑊1 could suffice and could perhaps reduce computational
𝑅𝐴 (𝑚), complexity with marginal loss in performance. In order to
When 𝑘 = 𝑛𝑠𝑐
make frequency offset estimation independent of 𝜏, we
𝑌𝑚,𝑖 construct,
𝑅𝐴 𝑊 = |𝑊1 | × 𝑊1 − 𝑗|𝑊2 | × 𝑊2
ℎ𝑚,𝑖 𝑒 𝑗2𝜋𝜉(𝑚𝑁𝑔+𝑖𝑁−𝜏) 𝑒 −𝑗2𝜋𝑛𝑠𝑐 (𝑚)𝜏⁄𝑁 (1 − 𝑒 𝑗2𝜋𝜉𝑁 ) (4)
−𝑗2𝜋𝜉𝑁𝑔
= . ∝𝑒 (cos 2 (2𝜋𝜏 ⁄𝑁) + sin2 ((2𝜋𝜏 ⁄𝑁))
1 − 𝑒 𝑗2𝜋𝜉
∝ 𝑒 −𝑗2𝜋𝜉𝑁𝑔 (10)
There will be energy on other subcarriers as well. But the The frequency offset 𝜉, can then be estimated by computing the
leakage is assumed to be small. Combining within a symbol angle of 𝑊 as,
group m, and assuming that the channel is invariant within a 1
𝜉=− 𝑎𝑟𝑐𝑡𝑎𝑛(𝑊) . (11)
symbol group, we get (per antenna), 2𝜋𝑁𝑔
4
Note that the maximum frequency offset which can be
𝑌𝑚 = ∑ 𝑌𝑚,𝑖 estimated with the above method corresponds to, 𝜉𝑚𝑎𝑥 =
𝑖=0 (5) 1 1
𝑅𝐴 ± =± ≈ 357𝐻𝑧, which is higher than the
ℎ𝑚 𝑒 𝑗2𝜋𝜉(𝑚𝑁𝑔−𝜏) 𝑒 −𝑗2𝜋𝑛𝑠𝑐 (𝑚)𝜏⁄𝑁 (1 − 𝑒 𝑗2𝜋𝜉5𝑁 ) 2𝑁𝑔 2∗1.4𝑚𝑠
= . frequency offset setting of 200Hz for RAN4 specified
1 − 𝑒 𝑗2𝜋𝜉
Noise within a symbol group 𝑚 can be computed as, NPRACH missed detection requirement (see table 8.5.3.2.1-1
2 2 of [14]). Note that the frequency offset need not be explicitly
𝑁𝑚 = |𝑌𝑚,0 − 𝑌𝑚,1 | + |𝑌𝑚,3 − 𝑌𝑚,4 | (6) computed. The phaser corresponding to the frequency offset
The problem is to estimate the two unknowns, frequency offset can be computed as,
𝜉 and RTD 𝜏 using all available 𝑌𝑚 .

2327-4662 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2019.2957641, IEEE Internet of
Things Journal
4

𝑊 (12)
= 𝑒 −𝑗2𝜋𝜉𝑁𝑔 .
|𝑊|
Correcting the estimated frequency offset we get,
𝑉𝑚 = 𝑍𝑚 𝑒 𝑗2𝜋𝜉𝑁𝑔 ∝ 𝑒 𝑗2𝜋𝐻𝑚𝜏⁄𝑁 and (13)
𝑉3 = 𝑍3 𝑒 −𝑗2𝜋3𝜉𝑁𝑔 ∝ 𝑒 𝑗2𝜋𝐻3𝜏 ⁄𝑁

B. Delay estimation
Defining,
𝑉 , 𝑖𝑓 𝐻4𝑟+𝑙 = 𝐻𝑙 (14)
𝑈4𝑟+𝑙 = { 4𝑟+𝑙
∗
𝑉4𝑟+𝑙 , 𝑖𝑓 𝐻4𝑟+𝑙 = −𝐻𝑙
with 𝑙 ∈ {−1, +1, 6} the possible hopping sequences and
summing across repetitions 𝑟, we get,
𝑇 = ∑ 𝑈 ∝ 𝑒 𝑗2𝜋𝛼𝑙𝜏 ⁄𝑁 (15) Figure 2-3: Polynomial interpolation around FFT bin with maximum
𝑙 4𝑟+𝑙
value[16].
𝑟=0,1,…𝑅
Modify 𝑇𝑙 such that, 𝛼0 = −1, 𝛼1 = 6, 𝛼2 = 1 and 𝛼3 = −6. C. Signal detection
For a given NPRACH opportunity (parameterized by the
𝑅𝐴 (0)) NPRACH or DTX detection is determined based on comparing
subcarrier of the first symbol group 𝜌 = 𝑛𝑠𝑐 form vector
the metric 𝐺 ∗ normalized with the noise averaged across
𝐹𝜌 = [−𝑇3 0 0 0 0 𝑇0 0 𝑇2 0 0 0 0 𝑇1 0 … … 0] with size 𝑁𝜏
symbol groups, repetitions and antennas with a threshold
being the FFT size used for 𝜏 estimation. The RTD can be 𝑅
𝜖𝑁𝑃𝑅𝐴𝐶𝐻 . 𝑅
Here 𝜖𝑁𝑃𝑅𝐴𝐶𝐻 is a constant determined
estimated using ML estimator based on FFT. Let, experimentally by sending a zero signal (DTX) and computing
𝑁𝜏 −1
−𝑗2𝜋𝑠 the normalized value (normalized with noise power) of 𝐺 ∗ over
𝑄𝑣 = ∑ 𝐹𝜌 (𝑠)𝑒 𝑁𝜏 and several DTX frames. The cumulative distribution function
𝑠=0 (16) (CDF) of the observed values of 𝐺 ∗ gives the probabilities of
𝐺𝑣 = ∑|𝑄𝑣 |2 the values of 𝐺 ∗ . The threshold 𝜖𝑁𝑃𝑅𝐴𝐶𝐻
𝑅
is determined from the
∗
𝑁𝑟𝑥 CDF by finding the value of 𝐺 occurring with a probability of
The timing hypothesis vector 𝐺𝑣 corresponding to the 99.9% (corresponding to the false alarm rate of 0.1%). The
maximum absolute value is the RTD estimate. Specifically, detection on a subcarrier 𝜌 is done as follows.
[𝑚𝑚𝑎𝑥 , 𝑖𝑑𝑥𝑚𝑎𝑥 ] = argmax{𝐺𝑣 }. (17)
[0,𝑁𝜏 ]
RTD estimate can be computed using the FFT bin index 𝐺∗
Let 𝜆 = 1 . (18)
(𝑖𝑑𝑥𝑚𝑎𝑥 ) corresponding to the maximum value 𝑚𝑚𝑎𝑥 as, 𝜏̃ = ∑𝑅 ∑3 𝑁
𝑁𝑟𝑥 4𝑅 𝑟=0 𝑚=0 𝑚
𝑖𝑑𝑥𝑚𝑎𝑥 𝑅𝐴 (𝜌)
. Using 𝑁𝜏 = 256 will give a resolution of 𝐷𝑒𝑡𝑒𝑐𝑡𝑖𝑜𝑛 𝑜𝑛 𝑛𝑠𝑐
𝑁𝜏 ∗(3.75)𝑘𝐻𝑧
1 𝑁𝑃𝑅𝐴𝐶𝐻, 𝑖𝑓
𝑅 (19)
(3.75)𝑘𝐻𝑧
= 1.04 𝑢𝑠. Given that the timing offset will be within ={ λ ≥ 𝜖𝑁𝑃𝑅𝐴𝐶𝐻
256
the CP and the timing error is ±7 × 16𝑇𝑠 , the maximum is 𝐷𝑇𝑋, 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒
taken after zeroing out the FFT output corresponding to timing
offsets outside the window determined by CP size and the For the subcarrier 𝜌 on which NPRACH is detected, 𝑆𝑁𝑅𝜌 = 𝜆
timing error. A quadratic interpolation [16] around the can be reported to upper layers for scheduling/uplink power
maximum value provides better resolution in the timing control algorithms.
estimate (but with additional computational complexity). Let
the values of 𝐺𝑣 at 𝑚𝑚𝑎𝑥 − 1 , 𝑚𝑚𝑎𝑥 and 𝑚𝑚𝑎𝑥 + 1 be 𝛼 , 𝛽 3. NPUSCH FORMAT 1 (DATA) RECEIVER
and 𝛾 respectively as show in Figure 2-3. The maximum value NB-IoT UL shared channel data is mapped to NPUSCH
after thresholding is given by, Format 1 and supports 1,3,6, or 12 tones at 15 kHz subcarrier
1 spacing or a single tone at 3.75 kHz subcarrier spacing. Since
𝐺 ∗ = 𝑦(𝑝) = 𝛽 − (𝛼 − 𝛾) × 𝑝, the bandwidth is limited to 180 kHz, resource units (RU) are
4
1
(𝛼−𝛾) defined, which are essentially resource blocks allocated in time.
2
where 𝑝 = . RTD estimate can be refined after Details on resource unit duration in millisecond (ms), slot
𝛼−2𝛽+𝛾
𝑖𝑑𝑥(𝑚𝑎𝑥) +𝑝 lengths in ms, modulation used, number of symbols and
interpolation using 𝜏̃ = . For 128 repetitions5, FFT references signals are shown for NPUSCH Format 1 in Figure
𝑁𝜏 ∗(3.75)𝑘𝐻𝑧
at the output of two receive antennas and groups of 64 3-1. NPUSCH Format 1 has one pilot symbol per slot, uses
𝜋 𝜋
repetitions are non-coherently combined before determining the Turbo coding and uses either QPSK or BPSK modulation.
4 2
maximum.

5
Based on the link budget analysis for NPRACH in small cells, high
repetitions may not be necessary for the targeted MCL of 164dB.

2327-4662 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2019.2957641, IEEE Internet of
Things Journal
5

NPUSCH Format 1 is used to carry uplink data, which is

processed in the form of transport blocks (TB). TD Samples @ fs STO
Remove CP/ Correction Pilots
For 3.75KHz, Mod. is pi/2 FFT demodulation
3.75KHz SCS

(multi-tone)
NPUSCH Format 1 (Data)
48 tones

BPSK or pi/4 QPSK

Frequency
For 3.75KHz, 1RU = 32ms Freq.offset
smoothing
2ms slots estimation
(multi-tone)
slot length = For 15KHz, RU size Modulation for
RU = 2ms
0.5ms varies with #tones. 15KHz Channel
1ms 12 tones Estimation Freq.
1ms=2x0.5slots Equalized Offset
180 KHz = 12 tones

MRC
14 symbols 6 tones QPSK Symbols/
2 RS Weights Corrected
15KHz SCS

12 Data 3 tones LLRs Noise Pilots

1 tone Estimation
pi/2 BPSK NPUSCH DTX or
or pi/4
Format 2 Detection ACK/NACK
NPUSCH
QPSK Descramble/ Decoded
Format 1 Turbo Decode bits
De-interleave

RU =1 ms RU = 4 ms
RU = 8 ms Figure 3-2:Receiver blocks for NPUSCH F1 and F2.
Figure 3-1:NPUSCH Format 1 (Data) Resource Unit Structure. To minimize the phase discontinuity between symbol
A transport block is encoded as one coding block and boundaries, a modulation-based phase rotation is applied (see
𝑁𝑃𝑈𝑆𝐶𝐻 section of 10.1.5 of [10]). For symbol l of a continuous
transmitted 𝑁𝑟𝑒𝑝 times contiguously, unless uplink
𝑁𝑃𝑈𝑆𝐶𝐻 transmission over subcarrier 𝑘, the modulation-based phase
transmission gaps are required. Each of the 𝑁𝑟𝑒𝑝 repetitions
rotation to be applied is 𝑒 𝑗𝜙𝑘,𝑙 . The modulation-based phase
is mapped to 𝑁𝑅𝑈 resource units (RU), where 𝑁𝑅𝑈 may be one 𝜋 𝜋
rotation essentially consists of two parts, first part is the or
of the values in {1, 2, 3, 4, 5, 6, 8, 10} depending on the TBS 2 4
size and MCS index (see section 16.5.1.2 of [13] for TBS table). phase rotation for BPSK or QPSK modulation respectively
A resource unit is defined as 𝑁𝑠𝑦𝑚𝑏 𝑈𝐿 𝑈𝐿
𝑁𝑠𝑙𝑜𝑡𝑠 consecutive SC- which is applied to odd-numbered symbols, second the
𝑅𝑈 accumulative phase rotation part that is used to ensure the
FDMA symbols in the time domain and 𝑁𝑆𝐶 consecutive
𝑅𝑈 𝑈𝐿 desired phase difference between symbol boundaries. This
subcarriers in the frequency domain, where 𝑁𝑆𝐶 and 𝑁𝑠𝑦𝑚𝑏 =
phase rotation is accumulative throughout a contiguous
7 are the number of assigned tones and number of symbols in a transmission. The received signal for slot 𝑠 and symbol 𝑙 can be
𝑁𝑃𝑈𝑆𝐶𝐻
slot respectively (see [11]). The 𝑁𝑟𝑒𝑝 repetitions are written as,
𝑁𝑃𝑈𝑆𝐶𝐻
𝑁𝑟𝑒𝑝
grouped into 𝑁𝑃𝑈𝑆𝐶𝐻 repetition cycles. Within each cycle, 𝑦𝑠,𝑙 [𝑛] = ℎ𝑠,𝑙 𝑒 𝑗2𝜋𝜉(𝑛−𝛿𝑡) 𝑥𝑠,𝑙 [𝑛 − 𝛿𝑡] + 𝑤𝑠,𝑙 [𝑛]
𝑀𝑖𝑑𝑒𝑛𝑡𝑖𝑐𝑎𝑙
𝑗2𝜋𝑘
(𝑛−𝛿𝑡)
(20)
𝑈𝐿 𝑁𝑃𝑈𝑆𝐶𝐻
every 𝑁𝑠𝑙𝑜𝑡𝑠 are repeated 𝑀𝑖𝑑𝑒𝑛𝑡𝑖𝑐𝑎𝑙 − 1 identical times before = ℎ𝑠,𝑙 𝑒 𝑗2𝜋𝜉(𝑛−𝛿𝑡) 𝛼𝑙 𝑒 𝑗𝜙𝑘,𝑙 𝑒 𝑁𝑙 + 𝑤𝑠,𝑙 [𝑛].
𝑈𝐿
the next 𝑁𝑠𝑙𝑜𝑡𝑠 are transmitted, where where 𝜉 is the frequency offset between the UE and eNB
𝑁𝑃𝑈𝑆𝐶𝐻
𝑀𝑟𝑒𝑝
𝑁𝑃𝑈𝑆𝐶𝐻
𝑀𝑖𝑑𝑒𝑛𝑡𝑖𝑐𝑎𝑙 = min ( , 4) 𝑅𝑈
𝑖𝑓 𝑁𝑆𝐶 >1 and transmit/receive frequencies normalized with sampling
2 frequency; 𝛿𝑡 is the sampling time offset normalized with
𝑁𝑃𝑈𝑆𝐶𝐻 𝑅𝑈
𝑀𝑖𝑑𝑒𝑛𝑡𝑖𝑐𝑎𝑙= 1 𝑖𝑓 = 1. 𝑁𝑆𝐶 symbol duration; ℎ𝑠,𝑙 is the channel coefficient at the 𝑙 𝑡ℎ symbol
Consider the 𝑁𝑙 samples from 𝑠 𝑡ℎ slot and the 𝑙 𝑡ℎ symbol i.e. in of 𝑠 𝑡ℎ slot; 𝑤𝑠,𝑙 is the white Gaussian noise at the 𝑠 𝑡ℎ slot and
𝑙
the range 𝑛 = 𝑁𝑠,𝑙 − 𝑁𝑐𝑝 , … , 𝑁𝑠,𝑙 + 𝑁𝑙 − 1. Here 𝑁𝑠,𝑙 are the
𝑙
𝑙 𝑡ℎ symbol and 𝑗 = √−1. Dropping the CP and taking FFT,
samples up to this symbol given by, 𝑁𝑠,𝑙 = 𝑠𝑁𝑠𝑙𝑜𝑡 + ∑𝑙 (𝑁𝑐𝑝 + we get equation frequency domain samples as,
𝑗2𝜋𝑘𝑛
𝑁𝑠,𝑙 +𝑁𝑙 −1
𝑁𝑙 ). Let, 𝑥𝑠,𝑙 [𝑛] = 𝛼𝑙 𝑒 𝜙𝑘,𝑙 𝑒 𝑁𝑙 is the transmit signal on the −
𝑗2𝜋𝑛𝑘

𝑘 𝑡ℎ subcarrier with modulated alphabet 𝛼𝑙 . Here, 𝑁𝑠𝑙𝑜𝑡 is the 𝑌𝑠,𝑙 [𝑘] = ∑ 𝑦𝑠,𝑙 [𝑛]𝑒 𝑁𝑙 .
(21)
size of a 0.5ms slot6. 𝑁𝑙 is the size of a symbol, with 𝑙 ∈ 𝑛=𝑁𝑠,𝑙
𝑙
{0, … ,6}. 𝑁𝑐𝑝 is the cyclic prefix7 size of the 𝑙 𝑡ℎ symbol. 𝑥𝑠,𝑙 [𝑛]
Simplifying further we get (adding 𝛿𝑡 notation to indicate
is the time domain waveform at 𝑛𝑡ℎ sample of 𝑙 𝑡ℎ symbol in 𝑠 𝑡ℎ
dependence on timing offset8), the frequency domain sample
slot. Also, 𝜙𝑘,𝑙 is the phase of the 𝑘 𝑡ℎ subcarrier.
for slot 𝑠, symbol 𝑙 , on the 𝑘 𝑡ℎ tone as,

6 8
This is equal to 960 samples at sampling rate of 𝑓𝑠 = 1.92 𝑀𝑠𝑝𝑠. The frequency domain samples of the received signal are also a function of
7
At 𝑓𝑠 = 1.92 𝑀𝑠𝑝𝑠,this corresponds to 𝑁𝑐𝑝 𝑙 𝑙
= 10 𝑓𝑜𝑟 𝑙 = 0 and 𝑁𝑐𝑝 = the frequency offset 𝜉, but the notation is dropped to avoid clutter.
9 𝑓𝑜𝑟 𝑙 > 0.

2327-4662 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2019.2957641, IEEE Internet of
Things Journal
6

𝛿𝑡 2
𝑌𝑠,𝑙,𝑘 𝑚𝑠,𝑙,𝑘 , since |𝑟𝑠,𝑙,𝑘 | = 1. An FFT based frequency offset
−𝑗2𝜋𝑘𝛿𝑡 estimator is generally used on such phase ramped signals. For
ℎ𝑠,𝑙,𝑘 𝛼𝑙 𝑒 𝑗𝜙𝑘,𝑙 𝑒 𝑗2𝜋𝜉(𝑁𝑠,𝑙−𝛿𝑡) 𝑒 𝑁𝑙 (1 − 𝑒 𝑗2𝜋𝜉𝑁𝑙 ) NPUSCH Format 1, 𝑙 = 3 is the pilot symbol. For multi-tone
= (22)
(1 − 𝑒 𝑗2𝜋𝜉 ) transmissions, frequency smoothing is first applied to the
+ 𝑊𝑠,𝑙,𝑘 . demodulated pilot. A linear average of all tones in each of the
received SC-FDMA symbol it taken to increase SNR under
relatively flat channel conditions. Specifically,
A. Timing offset estimation for multi-tone assignments:
In general, the received signal is not time aligned and typically
has a non-zero sampling time offset. Timing offset estimation 𝑅𝑈
𝑁𝑆𝐶 −1
is needed in the receiver because different tones have different 1 (26)
phases which can generally be captured by the channel estimate ℎ̂𝑠,3 = 𝑅𝑈 ∑ ℎ̂𝑠,3,𝑘
𝑁𝑆𝐶
𝑘=0
as well. However, to enhance performance, it is better to 𝑅𝑈
Here 𝑁𝑆𝐶 is the number of subcarriers of the resource unit
separate this offset from channel estimate. In the case of single
assigned to the user under consideration. The averaged signal is
tone, this is not needed because there is no second tone, so the
used for channel estimation, frequency offset estimation and for
problem of different phases between tones does not even arise.
noise power estimation. The pilot symbols (after conjugate
Computing the phase difference across adjacent subcarriers 𝑘
matching and averaging because of flat channel in the narrow
and 𝑘 + 1, we get,
𝑗2𝜋𝛿𝑡 band) form a noisy phase ramp signal with a sampling period of
𝛿𝑡
𝑃𝑘 = 𝑌𝑠,𝑙,𝑘 𝛿𝑡
∗ conj(𝑌𝑠,𝑙,𝑘+1 ) ∝ 𝑒 𝑁𝑙 . (23) 0.5 ms.
Summing the phase differences across subcarriers for a give slot C. Frequency offset estimation:
𝑁𝑠𝑐 −1
𝑠, we get 𝑄𝑘𝑠 = ∑𝑘=1 𝑃𝑘 . The sampling time offset for a given
slot can be estimated as, For frequency offset and channel estimation a block-based
1 processing scheme is adopted to avoid large memory
̃ =
𝛿𝑡 arctan(𝑄𝑘𝑠 ). (24)
2𝜋𝑁𝑙 requirements during higher repetitions and to obtain processing
The average sampling time offset across 𝑛 slots can be used gain during low operating SNR. Block sizes of 𝐵 = 8 𝑚𝑠 is
for final sampling time offset correction. Note that after the used and block size of 𝐵 = 32 𝑚𝑠 for multi and single tone
initial NPRACH access, UE would apply the eNB suggested assignments respectively, when a single transmission time are
timing advance (within a ±3.43𝜇𝑠 accuracy [14]) to achieve greater than 𝐵, otherwise the transmission time is used as the
uplink synchronization and whatever subsequent sampling time block size. The block size is chosen such that the RAN4
offset which is seen by eNB is either due to the residual timing performance targets are met. The block size is chosen to be
offset or due to the drift in the UEs local oscillator over time. If 0.423
smaller than the channel coherence time ≈ = 423 ms for
the UE timing off by a large value, an NPDCCH order is 𝑓𝑑

typically sent to allow the UE to reconnect. Note that sampling 𝑓𝑑 = 1𝐻𝑧 Doppler. With pilots occurring every 0.5ms and
time offset need not be computed explicitly, and the sampling using a 𝑁 = 256 point FFT, we get a resolution of
time offset correction on the received tones can be done using 2000 𝐻𝑧⁄256 ≈ 7.8 𝐻𝑧 in frequency offset 𝜉. The output of
̂
𝑗2𝜋 𝛿𝑡 the FFT is squared and summed over the receive antennas and
𝑄𝑘𝑠 𝑁𝑙
the phasor, =𝑒 . The timing correction can be applied the index corresponding to the maximum value provides an
|𝑄𝑘𝑠 |
𝑄𝑘𝑠 estimate of the frequency offset as shown in equation (26).
𝛿𝑡
as 𝑌𝑠,𝑙,𝑘 = 𝑌𝑠,𝑙,𝑘 ∗ . 𝑁−1
|𝑄𝑘𝑠 | −𝑗2𝜋𝑚
𝐻𝑣 = ∑ ℎ̂𝑠,3 𝑒 𝑁

𝑚=0
B. Pilot Demodulation:
𝐼𝑣 = ∑|𝐻𝑣 |2
The first step is pilot demodulation i.e., removal of known (27)
𝑁𝑟𝑥
values of the pilots from the received pilot signal by a 𝑚𝑚𝑎𝑥 = argmax{𝐼𝑣 }
multiplication of the signal with the conjugated version of the 𝑁𝑁
[− , ]
4 4
known pilot sequence. For 𝑘 𝑡ℎ tone of the 𝑙 𝑡ℎ symbol in slot 𝑠
(all within a block 𝑏), pilot demodulation can be written as, The maximum is taken over a range of possible frequency offset
ℎ̂𝑠,𝑙,𝑘 = 𝑟𝑠,𝑙,𝑘
∗
𝑌𝑠,𝑙,𝑘 (25) values of [−250 250]𝐻𝑧. Further accuracy in the estimate can
Detailed information on generation of the known pilot sequence be obtained by interpolating around the maximum value (like
𝑟𝑠,𝑙,𝑘 can be found in [S10]. Let the transmitted pilot signal be what was defined in NPRACH receiver section 2.B) as follows.
𝑟𝑠,𝑙,𝑘 , the frequency offset be 𝜉 (normalized with sampling Let the values of 𝐺𝑣 at 𝑚𝑚𝑎𝑥 − 1 , 𝑚𝑚𝑎𝑥 and 𝑚𝑚𝑎𝑥 + 1 be 𝛼
period) and noise be 𝑛𝑠,𝑙,𝑘 , then the received pilot signal , 𝛽 and 𝛾 respectively. The offset 𝑝 of the true peak w.r.t 𝑚𝑚𝑎𝑥
1
(assuming that the received signal is time offset corrected) can (𝛼−𝛾)
is given by, 𝑝 = 2
. The frequency offset estimate 𝜉̂ is
be written as 𝑌𝑠,𝑙,𝑘 = ℎ𝑠,𝑙,𝑘 𝑟𝑠,𝑙,𝑘 𝑒 𝑗2𝜋𝜉𝑁𝑠,𝑙 + 𝑛𝑠,𝑙,𝑘 . Multiplying 𝛼−2𝛽+𝛾
𝑚𝑚𝑎𝑥 +𝑝
with the complex conjugate 𝑟𝑠,𝑙,𝑘 ∗
results in ℎ𝑠,𝑙,𝑘 𝑒 𝑗2𝜋𝜉𝑁𝑠,𝑙 + given by, 𝜉̂ = .
𝑁𝑁𝑠𝑙𝑜𝑡

2327-4662 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2019.2957641, IEEE Internet of
Things Journal
7

D. Channel estimation: belonging to the same resource units) are combined before
The channel estimation process is repeated every block for descrambling and de-interleaving. Rate de-matching is
each active UE. The channel estimates are averaged over the performed for the de-interleaved LLRs for each repetition and
block. This design follows the assumption of a slow fading the repetitions of the same transmission are combined with
channel (≈ 1Hz Doppler frequency). The time averaging HARQ LLRs of earlier transmissions if available. For every
operation is done together with the frequency offset correction non-identical repetition, a turbo decoding operation is
operation for antenna 𝑟 as, performed followed by a CRC check. Bits outputs from the
2𝐵−1
(28) decoder after successful decoding, including CRC check
1 ̂
ℎ̂𝑏,𝑟 = ∑ ℎ𝑠,3 𝑒 −𝑗2𝜋𝜉 𝑁𝑠,3 . indication for each identical repetition group are transferred to
2𝐵 higher layer. The SNR over a block 𝑏, is given by,
𝑠=0
For multiple receive antennas, ℎ̂𝑏,𝑟 is a vector of size 𝑁𝑟𝑥 with 𝑆𝑁𝑅𝑏 = 𝑤𝑏 ℎ̂𝑏 . (35)
elements corresponding to channel estimates at each of the The SNR reported to higher layers is generally the average SNR
receive antennas. Note that the subscript 𝑏 denotes a given over the last few blocks to avoid long time averaging for higher
block. The frequency offset corrected data symbols for antenna repetitions.
𝑟 are given by,
̂ 4. NPUSCH FORMAT 2 (CONTROL) RECEIVER
𝑑𝑠,𝑙,𝑘,𝑟 = 𝑌𝑠,𝑙,𝑘 𝑒 −𝑗2𝜋𝜉 𝑁𝑠,𝑙 (29)
The noise variance is computed as shown below. The NPUSCH Format 2 channel carries acknowledgement of
𝜎2𝑏 = downlink data and supports only single tone with either 3.75
2𝐵−1 kHz or 15 kHz subcarrier spacing. The NPUSCH Format 2
1 2 (30) channel has a resource unit (RU) duration of 2 ms and has three
∑ ∑ ∑ |ℎ̂ 𝑠,3,𝑘,𝑟 𝑒−𝑗2𝜋𝜉̂𝑁𝑠,3 − ℎ̂ 𝑏,𝑟 |
2𝐵𝑁𝑟𝑥 𝑁𝑅𝑈
𝑠𝑐 reference symbols per slot to facilitate robust estimations while
𝑁𝑟𝑥 𝑁𝑅𝑈
𝑠𝑐 𝑠=0 using a single 3.75 kHz/15 kHz tone. In this work, only the 15
Note that there are two slots per subframe and hence the kHz sub-carrier spacing is considered for NPUSCH Format 2.
summation is over 2𝐵 for a block of size 𝐵 ms. When the block The single tone NPUSCH Format 2 is similar to the single tone
size 𝐵 is small, the number of samples in the noise estimate is NPUSCH Format 1, except for the pilot structure, channel
too small and the estimates could be biased. In such a scenario, coding and modulation scheme.
the SNR could be estimated as,
NPUSCH Format 2 (Control)
𝜎𝑏2 =
3.75kHz SCS

2𝐵−1 For 3.75KHz

1
𝑅𝑈
∑ ∑ ∑ |ℎ̂𝑠,3,r 𝑒 −𝑗2𝜋𝜉𝑁𝑠,3 − ℎ̂𝑏 |.2 (31) 1RU =8ms
2𝐵𝑁𝑟𝑥 𝑁𝑠𝑐 𝑅𝑈 𝑠=0
𝑁𝑟𝑥 𝑁𝑠𝑐
2ms slots
1 Mod. is pi/2 BPSK for 3.75KHz
Here ℎ̂𝑏 = ∑2𝐵−1 ℎ𝑠,3 𝑒 −𝑗2𝜋𝜉𝑁𝑠,3 (time averaged) and ℎ̂𝑠,3,r =
2𝐵 𝑠=0
1 𝑅𝑈
𝑁 −1 For 15KHz
∑ 𝑆𝐶 ℎ̂𝑠,3,𝑘 (frequency averaged). Also, ℎ̂𝑠,3,𝑘 is the slot 1ms = 2 slots
𝑁𝑅𝑈 𝑘=0
𝑆𝐶
demodulated, STO corrected (multi-tone) reference symbol in length = 14 symbols Modulati
15kHz SCS

a given slot 𝑠 and for a given tone 𝑘. Under MMSE criteria, 0.5ms 6 RS on is pi/2
4 slots 8 Data BPSK for
equalizer taps for block 𝑏 and antenna 𝑟 are calculated in the
RU=2ms 15KHz
following manner,
𝑤𝑏,𝑟 = 𝜎𝑏−2 ℎ̂𝑏,𝑟
𝐻
, (32) Single tone
𝐻
with (⋅) denoting the Hermitian operation. The equalized 1RU
symbols in a block 𝑏 are given by, Figure 4-1:NPUSCH Format 2 (Control) Resource Unit
Structure.
𝑒𝑠,𝑙,𝑘 = ∑ 𝑤𝑏,𝑟 𝑑𝑠,𝑙,𝑘,𝑟 . (33)
𝑁𝑟𝑥
where the subscript 𝑟 denotes the receive antenna. The inverse NPUSCH Format 2 has three pilot symbols per slot, uses
𝑅𝑈
Fourier transform of order 𝑁𝑠𝑐 ∈ {3,6,12} is computed as, repetition coding on one ACK/NACK bit and uses only a
𝜋
𝑅𝑈 −1
𝑁𝑠𝑐 2
𝑗2𝜋𝑛𝑘
𝑡
1 − 𝑅𝑈 BPSK modulation scheme. The frequency offset estimation and
𝑒𝑠,𝑙,𝑘 (𝑛) =
∑ 𝑒𝑠,𝑙,𝑘 𝑒 𝑁𝑠𝑐 (34)
𝑅𝑈 channel estimation modules proposed in previous section for
√𝑁𝑠𝑐 𝑘=0 NPUSCH Format 1 can be reused for NPUSCH Format 2 also.
The log-likelihood ratios (LLRs) are obtained as, One minor change, however, is that in case of NPUSCH Format
𝑡 𝑡
2, there are three pilots available in a slot of 0.5ms duration
−𝑅𝑒𝑎𝑙(𝑒𝑠,𝑙,𝑘 ) − 𝐼𝑚𝑎𝑔(𝑒𝑠,𝑙,𝑘 ) for 𝐵𝑃𝑆𝐾 compared to a single pilot in case of NPUSCH Format 1. The
𝑙𝑙𝑟𝑠,𝑙,𝑘 = { 𝑡 𝑡
(−𝑅𝑒𝑎𝑙(𝑒𝑠,𝑙,𝑘 ), −𝐼𝑚𝑎𝑔(𝑒𝑠,𝑙,𝑘 )) for 𝑄𝑃𝑆𝐾 three pilots are time averaged, assuming minor phase changes
between adjacent symbols. Pilot demodulation can then
The LLRs corresponding to identical repetitions (i.e., LLRs proceed as in NPUSCH Format 1. Reusing NPUSCH Format 1

2327-4662 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2019.2957641, IEEE Internet of
Things Journal
8

modules has the advantage of saving memory in an obtained per millisecond per antenna and averaging over 𝐵 =
implementation, reduction in development and module testing 4𝑚𝑠 gives better estimation accuracy. Specifically, noise can
efforts. For NPUSCH Format 2, there is no decoding of the be estimated as,
LLRs involved and only the ACK/NACK detection has to be 2𝐵−1
2
1 2
designed, as discussed below. Let 𝛼𝑙 = 𝑡𝑙 𝑢 be the modulated 𝜎 = ∑ ∑ |𝑑𝑠,𝑙,𝑘0 − 𝑑𝑠,𝑙+1,𝑘0 | (39)
alphabet transmitted on sub-carrier 𝑘 = 𝑘0 , where 𝑡𝑙 is the 2𝐵𝑁𝑟𝑥
𝑁𝑟𝑥 𝑠=0
scrambling and 𝑢 is the unknown ACK/NACK bit. For every 𝑙=0,1,5,6

symbol, compute, where 𝑑𝑠,𝑙,𝑘0 is the unscrambled data symbol 𝑙 corresponding to

𝑍𝑠,𝑙 [𝑘0 ] = 𝑌𝑠,𝑙 [𝑘0 ]𝑡𝑙∗ 𝑒 −𝑗𝜙𝑘0,𝑙 , (36) 𝑠 𝑡ℎ slot transmitted on sub-carrier 𝑘0 . Note that there are two
slots per subframe and hence the summation is over 2𝐵 for a
where 𝑌𝑠,𝑙 [𝑘0 ] is defined in equation (21) and, 𝑒 𝑗𝜙𝑘0,𝑙 is the
block of size 𝐵 ms.
modulation-based phase rotation to be applied for symbol 𝑙 of
a continuous transmission over subcarrier 𝑘0 (see section of
5. IMPLEMENTATION ASPECTS
10.1.5 of [10]). Dropping the subcarrier notation 𝑘0 to avoid
𝑝
clutter, for slot 𝑠, we define variables 𝑍𝑠 , and 𝑍𝑠𝑑 for pilot and In a typical eNB implementation, concurrent processing of
data symbols respectively for each antenna as, legacy LTE, eMTC, NB-IoT uplink channels, is usually
p ̅
required with multiple users being served per transmit time
𝑍𝑠 = ∑ 𝑍𝑠,𝑙 ∝ ℎ𝑔 𝑒 𝑗2𝜋𝜉 𝑁𝑠,𝑙 interval (TTI). Therefore, it will be useful to discuss the various
𝑙=2,3,4 implementation aspects associated with NB-IoT uplink
̅
𝑍𝑠d = ∑ 𝑍𝑠,𝑙 ∝ ℎ𝑔 𝑢𝑒 𝑗2𝜋𝜉 𝑁𝑠,𝑙 receiver. In this section some of the important implementation
𝑙=0,1,5,6 aspects are discussed.

(𝑔+1)𝑃 (37) A. Common Phase Correction:

p −𝑗2𝜋𝜉̅ 𝑁𝑠,𝑙 In a practical setup, phase correction per symbol is required to
𝐽𝑟 (𝑢) = ∑ (𝑍𝑠 + 𝑢𝑍𝑠d )wb 𝑒
account for the difference between frequency location of the
𝑠=𝑔𝑃+1
𝑁𝑟𝑥 center of the Narrowband IoT physical resource block (PRB)
and the frequency location of the center of the LTE signal, since
𝐽(𝑢) = ∑|𝐽𝑟 (𝑢)|2 .
a single common wideband receiver FFT (of say size 2048 for
𝑟=1
20MHz bandwidth at 30.72Msps) is typically used in the digital
Here, 𝜉 ̅ is the estimated frequency offset (as shown in section
front end (DFE) for both legacy LTE and NB-IoT. Due to the
3.C), 𝑤𝑏 is the equalizer tap weight computed as in equation
common FFT, there will be a phase offset introduced for NB-
(32) but with noise variance computed differently as shown in
IoT tones due to the difference between center frequencies of
equation (39) of section 4.A. Also, 𝑔 represents the resource
legacy and NB-IoT. Consider the multi-tone NB-IoT signal
unit (RU) number, 𝑃 is the number of repetitions configured
𝑠(𝑡) (generated according as in section 10.1.5 in [11]),
and 𝑁𝑟𝑥 is the number of antennas. The sign of the equalized
symbol and the SNR of the transmission 𝑆𝑁𝑅𝑡 , which is the 5
1
sum of the SNRs of all blocks in the transmissions, i.e., 𝑆𝑁𝑅𝑡 = 𝑠(𝑡) = ∑ 𝑎𝑘 𝑒 𝑗2𝜋(𝑘+2)Δ𝑓(𝑡−𝑁𝑐𝑝,𝑙 𝑇𝑠) (40)
∑𝐵 𝑆𝑁𝑅𝑏 , are used as below to determine whether ACK, 𝑘=−6
NACK or DTX was transmitted,
𝐴 See Appendix A for symbol definitions. The origin in the above
𝑁 equation is at the beginning of the 𝑙 𝑡ℎ OFDM symbol (including
𝐷𝑇𝑋 , if 𝑆𝑁𝑅𝑡 ≤ 𝜖 (38)
cyclic prefix). Moving it to the beginning of the 𝑙 𝑡ℎ OFDM
= { 𝑠𝑖𝑔𝑛(argmax[𝑅𝑒(𝐽(𝑢) ]), 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒.
𝑢
symbol without CP, we get
𝐴 5
where, ∈ {1, 0, −1} represents the ACK/NACK or DTX 1
𝑁
𝑠(𝑡) = ∑ 𝑎𝑘 𝑒 𝑗2𝜋(𝑘+2)Δ𝑓𝑡 . (41)
information respectively. Here, 𝑆𝑁𝑅𝑏 is computed according to
𝑘=−6
equation (35), 𝜖 is a constant determined by sending a zero If the separation between the centers of LTE and NB-IoT
signal (DTX) and computing the value of 𝑆𝑁𝑅𝑡 over several carriers is 𝑓𝑜 = 𝑁𝑜 Δ𝑓, where Δ𝑓 is the sub-carrier spacing and
DTX frames. CDF of the observed values of 𝑆𝑁𝑅𝑡 is plotted to 𝑁0 is some constant, the received signal is given by,
determine the threshold 𝜖 which gives 1% DTX to ACK. 5 (42)
1
A. Noise estimation for NPUSCH F2: 𝑠(𝑡) = ∑ 𝑎𝑘 𝑒 𝑗2𝜋(𝑘+2)Δ𝑓𝑡 𝑒 𝑗2𝜋𝑓𝑜 (𝑡+𝑀𝑜 𝑇𝑠) .
𝑅𝑈 𝑘=−6
Since 𝑁𝑆𝐶 = 1 (i.e., single tone) for NPUSCH Format 2 and
Here, 𝑀𝑜 𝑇𝑠 is the time offset between beginning of frequency
the block size 𝐵 = 4𝑚𝑠, the noise estimation used in NPUSCH
shift application and the 𝑙 𝑡ℎ OFDM symbol (excluding CP)
Format 1 may lead to biased estimates during low SNRs due to
under consideration. Before taking FFT, the signal must be
insufficient averaging. Exploiting the fact that for NPUSCH 1
Format 2 all data symbols are identical i.e., either ACK or frequency shifted by ± Δ𝑓 (half tone shift) so that the FFT
2
NACK bit, the differences between adjacent symbols will output has some scaled version of the transmit tones (instead of
unmask the noise. With this approach, 4 samples can be frequency interpolated samples),

2327-4662 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2019.2957641, IEEE Internet of
Things Journal
9

5
1 1
)
𝑠(𝑡) = ∑ 𝑎𝑘 𝑒 𝑗2𝜋(𝑘+2)Δ𝑓𝑡 𝑒 𝑗2𝜋(𝑓𝑜 ±2Δ𝑓)(𝑡+𝑀𝑜 𝑇𝑠 (43)
Keysight
𝑘=−6 NB-IOT
Sampling gives, Signal
Generator
5
1 1
)
𝑠𝑛 = ∑ 𝑎𝑘 𝑒 𝑗2𝜋(𝑘+2)Δ𝑓𝑛𝑇𝑠 𝑒 𝑗2𝜋(𝑓𝑜 ±2Δ𝑓)(𝑛𝑇𝑠+𝑀𝑜 𝑇𝑠
(44)
𝑘=−6

Taking FFT results in,

𝑁−1
1 𝑗2𝜋𝑚𝑛 Device
𝑆𝑚 = ∑ 𝑠𝑛 𝑒 − 𝑁 (45) Under Test
𝑁
𝑛=0
𝑁−1 5 Spirent
1 1 1
)
𝑗2𝜋𝑚𝑛 Channel To/From
= ∑ ∑ 𝑎𝑘 𝑒 𝑗2𝜋(𝑘+2)Δ𝑓𝑛𝑇𝑠 𝑒 𝑗2𝜋(𝑓𝑜 ±2Δ𝑓)(𝑛𝑇𝑠+𝑀𝑜 𝑇𝑠 𝑒 − 𝑁 Emulator PC
𝑁
𝑛=0 𝑘=−6
1 Figure 6-1: Test Setup
𝑆𝑘+𝑁 1 1 = 𝑒 𝑗2𝜋(𝑁0±2)𝑀0𝑇𝑠Δ𝑓 𝑎𝑘+1±1
𝑜 +2±2 2 2 offset and channel estimations) with a block of size ≤ to 8/32
1 𝑚𝑠. The processing can be resumed after NPRACH
𝑗2𝜋(𝑁0 ± )𝑀0
2
= 𝑒 𝑁 𝑎𝑘+1±1 . transmission with a block of size ≤ 8/32 𝑚𝑠. From link level
2 2
simulations, it was observed that the impact on the performance
The phase correction to be applied at the NB-IoT receiver is
1 with the above proposed method was marginal.
𝑗2𝜋(𝑁0 ± )𝑀0
−( 2 )
𝑁
then 𝑒 . 6. PERFORMANCE RESULTS AND DISCUSSIONS

B. Uplink Transmission Gap Handling: In this section results are provided with link-level simulations
of all uplink channels and discussions are provided on the NB-
The frequency offset estimation and channel estimation and IoT receiver performance on a Qualcomm FSM Chipset based
identical LLR combing modules will be impacted in the LTE basestation. The test setup with which the performance of
presence of UL transmission gaps. From section 10.1.3.6 of NB-IoT uplink receiver was investigated is shown in Figure 6.1.
[10], any ongoing NPUSCH transmissions will be postponed by The practical setup was shown in Figure 6.1 consisted of a
keysight MXG signal generator capable of generating NB-IoT
the UE when an NPRACH resource overlaps in time and
uplink signals[19], a spirent channel emulator [20] box capable
frequency and will be resumed after the NPRACH opportunity
of generating 3GPP specified channel profiles (e.g., EPA
is complete. Also, to maintain time/frequency synchronization
channel with 5Hz Doppler and low correlation between
between UE and eNodeB during long UL repetition antennas), and the device under test (shown as MTP in the
transmissions, UL gaps can be created. During UL gaps, the UE Figure 6.1). The test procedure was performed as per [15] as
may switch to the DL and performs time/frequency follows. For the given NB-IoT Channel under test, do the
synchronization. If an NPUSCH UL transmission duration is ≥ following.
256 ms9, then a gap of 40 ms is inserted by the UE, followed Step 1) Adjust the AWGN generator, according to the
by the remaining transmissions. Further, UL gaps can be channel bandwidth (see section 8.5 of [15]. Step 2) The
created due to base-station configuring invalid subframes characteristics of the wanted signal shall be configured
during which NB-IoT transmission cannot take place [13][17]. according to the corresponding UL reference measurement
Such UL transmission gaps will require special handling by the channel defined in Annex A of [15]. Step 3) The multipath
receiver due to channel changing over the UL gap. There are fading emulators shall be configured according to the
several ways of handling this scenario. However, the simplest corresponding channel model defined in Annex B of [15]. Step
method could be to stop the block processing (for frequency 4) Adjust the frequency offset (if defined) of the test signal.

𝑵𝑻𝒙 𝑵𝑹𝑿 𝑵𝒓𝒆𝒑 Channel FO 𝝃 SNR[dB]

SNR at which NPRACH Performance NPRACH Performance
min detect. is for Format 0 with for Format 0 with Test
req. [14] Simulation Setup
Preamble Format
0
1 2 8 AWGN 0 -2.1 -4.5 -4.1
EPA1 Low 200 Hz 6.1 3.65 4.12
32 AWGN 0 -6.8 -9.23 -8.5
EPA1 Low 200 Hz 0.5 -1.75 -1.0
Table 1: NPRACH RAN4 performance summary.
9
For NPRACH, if the transmission duration is ≥ 64 × preamble duration,
then a gap of 40ms is inserted at the end of the NB-PRACH opportunity.

2327-4662 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2019.2957641, IEEE Internet of
Things Journal
10

Step 5) Adjust the equipment so that the SNR specified in Figure 6-3). The solid lines (black, red and blue) in Figures 6.2
Tables 8.5.1.5-2, 8.5.2.5-2 and 8.5.3.5-1 of [15] (for NPUSCH and 6.3 indicate the performance obtained with link level
Format 1, Format 2 and NPRACH respectively) is achieved at
the BS input. Step 6) The test signal generator sends a test
pattern (outlined in section 8.5 of [15]) with repetitions and the
receiver tries to detect the pattern. This pattern is then repeated.
For each of the fixed reference channels (FRC) (defined in
Table 8.5.1.5-3 of [15]) applicable to base station, the
throughput is measured (as per Annex E of [15]) for NPUSCH
Format 1 testing. For NPUSCH Format 2 testing, the statistics
of ACK/NACK/DTX is noted as per section 8.5.2.4.2 of [15].
For NPRACH, the statistic are collected as per section 8.5.3.4.2
of [15]. Only the RAN4 recommended tests [14] were studied
in both simulations and the test setups.
The RAN4 test results for NPRACH is summarized in Table
1 (see page 10). The SNR in dB column has three columns
under it. The first colum has the 3GPP specified minimum SNR
at which the probability of preamble detection is greater than or
equal to 99% with detection requirements as specified in section
8.5.3.2 of [14] and with false alarm probability being less than
or equal to 0.1% (see section 8.5.3.1 of [14]). The second Figure 6-2: Link Performance of NPUSCH F1 for single tone
column under SNR captures the NPRACH performance with
link simulations and the third column lists the NPRACH
performance on the test setup. As can be seen, there is a margin
of atleast >2dB ( ≈ 1.5 db) with link-level simulations and the
test setup over the RAN4 recommended SNR. This
demonstrates that the proposed NPRACH receiver comfortably
meets the RAN4 minimum requirements. Note that, becaue of
the multiplication between sum and difference differentials in
equation (10), the noise in the differentials will have a
multiplicative effect and the performance could be poorer with
AWGN channel or when the frequency offset is low. Using
𝑊 = 𝑊1 (in equation (10)) during such a scenario could
improve NPRACH performance. However, with NB-IoT
devices expected to be designed with low-cost hardware, there
could be a large frequency offset between the transmitter and
the receiver, and the step in equation (10) might become
necessary for better estimation of 𝜉.
The NPRACH performance obtained with the proposed Figure 6-3: Link Performance of NPUSCH F1 for 6 tones
design is comparable with the performance reported in [22].
The differences in link-level simulations and FSM simulations. The performance requirement of NPUSCH Format
mplementation (i.e., with test setup) is mainly due to the loss 1 is determined by a minimum required throughput for a given
from the analog receive filter which was not modelled in the SNR with HARQ retransmissions. The required throughput is
link level simulations, and the fixed point implemenation losses expressed as a fraction (70%) of the maximum throughput. The
(<0.5dB). Note that other impairments like antenna correlation, link-level simulation plots for the normalized throughputs
I/Q imbalance, phase noise, etc which were not modelled in the (normalized with the maximum throughput) are shown for
link level simulations, but could be part of the test setup. different NPUSCH Format 1 repetitions in Figure 6.2. It can be
Further, cabelling and signal splitter losses (within the Spirent seen that the demodulation performance improves (i.e., with
channel emulator shown in Figure 6.1) add to the losses in test higher margins) with higher repetitions, but incurs higher
setup. delays in detection/demodulation. However, for Small Cells,
The performance of NPUSCH Format 1 (data) for single and with coverage requirements being smaller than macro
6-tones are shown in Figure 6.2 and Figure 6.3 respectively. basestations [1], higher repetitions may rarely be configured.
Table 2 lists NPUSCH Format 1 test results (see page 11) for The summary of simulation and test setup results for all RAN4
all RAN4 specified (see section 8.5.1 of [14]) test cases. As per specified test cases with NPUSCH Format 1 using 15kHz
RAN4 [14], for single tone allocation, minimum performance subcarrier spacing (Table 8.5.1.1.1-3 of [14]) is captured in
is specified only for repetitions of 1, 16 and 64 (indicated by Table 2 (see next page). The results are in line with the
diamond markers in Figure 6.2), whereas for multi-tone performance reported in [23].
allocation, minimum performance is specified only for The minimum performance of NPUSCH Format 2 (control)
repetitions of 2, 16 and 64 (denoted by diamond markers in is summarized in Table 3. The ACK missed detection
probability shall not exceed 1% at the SNR given in table

2327-4662 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2019.2957641, IEEE Internet of
Things Journal
11

𝑵𝑻𝒙 𝑵𝑹𝑿 𝚫𝒇 Tones Channel FRC 𝑵𝒓𝒆𝒑 Fraction Minimum NPUSCH F1 NPUSCH F1
and of max. SNR at which performance performance
correlation throughput fraction of with with Test Setup
matrix throughput Simulation
should be
achieved
[14].
1 70% -2.1 -4.5 -4.2
ETU 1Hz A16-2
1 16 70% -8.8 -11.2 -11
Low
64 70% -12.6 -14 -13.8
2 70% -3.0 -7 -5
ETU 1Hz
3 A16-3 16 70% -8.1 -12 -11
Low
64 70% -11.4 -15.2 -14.5
1 2 15KHz
2 70% -0.6 -7 -3
ETU 1Hz
6 A16-4 16 70% -6.8 -13.5 -11
Low
64 70% -10.5 -16.5 -14
2 70% -0.7 -7 -3.5
ETU 1Hz
12 A16-5 16 70% -6.4 -14.5 -11
Low
64 70% -10.1 -17.5 -15
Table 2: RAN4 Results for NPUSCH Format 1(Data).

8.5.2.2.1-2 of [14] for 1Tx case. The ACK missed detection 1) NPRACH reliability and range enhancement:
probability is the probability of not detecting an ACK when an
ACK was sent per NPUSCH Format 2 transmission. The DTX The key enhancements made to NPRACH in 3GPP Release 15
to ACK probability for NPUSCH Format 2 case denotes the are mainly with regards to new preamble formats introduced for
probability that ACK is detected when nothing is sent on the range enhancements and time-division multiplexing (TDD)
wanted signal and only the noise is present per NPUSCH support. For FDD mode, there is one additional preamble
Format 2 transmission. The minimum requirement for DTX to format (i.e., Format 2 in frame-structure type 1) defined. For
ACK is defined in section 8.5.2.1.1 of [14] and shall not exceed TDD frame-structure type 2, two additional preambles formats
1% per NPUSCH Format 2 transmission. The proposed (0-a and 1-a) have been defined. A symbol group consists of a
NPUSCH Format 2 receiver meets the performance cyclic prefix of length 𝑇CP and a sequence of 𝑁 identical
requirements with a good margin (at least >0.9dB) with the test symbols with total length 𝑇SEQ . The total number of symbol
setup. The results are also in line with the performance reported groups in a preamble repetition unit is denoted by 𝑃. The
in [23]. If the implementational complexity is an issue for number of time-contiguous symbol groups is denoted by 𝐺.
NPUSCH Format 2, then only the data symbols can be Summary of preamble formats, cyclic prefix and sequence
considered in equation (37) for computing the detection metric, lengths are shown in Table 4 and Table 5 for FDD and TDD
with only a minor loss in performance. respectively. Also, the NPRACH transmission in 3GPP release
15 supports either a 3.75 kHz or 1.25 kHz sub-carrier spacing
A. Further enhancements to NB-IoT in Release 15.
(only with preamble Format 2 in FDD). The NPRACH formats
with different cyclic prefix, symbol group sizes and symbol
Several important enhancements have been added in Release 15 group repetitions have been added to cater for different ranges
for NB-IoT (see [25] and [26]) and in this section discussions and reliability. Symbol groups with Δ𝑓𝑅𝐴 =3.75 kHz (preamble
are provided on the key enhancements finalized in 3GPP Format 0/1/2 or 0-a/1-a) sub-carrier spacing hop by one or six
Release 15 for NB-IoT. The goal of this section is to show that sub-carriers in frequency as before, and symbol groups with
the enhancements to NB-IoT in Release 15 do not require any Δ𝑓𝑅𝐴 = 1.25 kHz (preamble Format 2 only) subcarrier spacing
significant changes to the receiver design proposed in this hop by one, three, or eighteen sub-carriers in frequency. In other
paper. words, frequency hopping within one NPRACH preamble is

𝑵𝑻𝒙 𝑵𝑹𝑿 Channel Tones 𝚫𝒇 𝑵𝒓𝒆𝒑 Minimum SNR in NPUSCH F2 NPUSCH F2

and dB at which performance with performance with
correlation detection Simulations Test Setup
matrix performance is
expected [14]
1 6.3 3.2 4.4
EPA 5Hz,
1 2 1 15KHz 16 -3.9 -5.8 -4.9
Low
64 -9.5 -11.8 -10.6
Table 3: RAN4 Results for NPUSCH Format 2(Control).

2327-4662 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2019.2957641, IEEE Internet of
Things Journal
12

defined for 3-levels of hopping with, -/+ 1.25 kHz, -/+ 3.75 kHz transmission (EDT) in uplink.
and 22.5 kHz (see [24]), with
• 1.25 kHz hopping gap for 1st to 2nd symbol group and • For non-EDT NPRACH resources, the UE postpones
with opposite direction for 5th to 6th symbol group. NPUSCH in those resources only if the UE indicates
• 3.75 kHz hopping gap for 2nd to 3rd symbol group and support for the corresponding feature(s).
with opposite direction for 4th to 5th symbol group. • For EDT NPRACH resources, the UE postpones
• 22. 5 kHz hopping gap for 3rd to 4th symbol group. NPUSCH in those resources only during an EDT
Repetitions of groups of symbol groups hop by a pseudo- procedure.
random number of sub-carriers in frequency. There are three in As discussed in section 5.B on the UL gap handling, the block
FDD, and four in TDD, possible cyclic prefix lengths for the processing for NPUSCH can be stopped and resumed with
random-access transmission symbol groups, suitable for partial block sizes. Non-coherent combining across partial
different maximum cell sizes (see version 15.3.0 of [10] for blocks could be adopted to improve the performance.
more details). The values for NPRACH periodicity are 40, 80,
160, 320, 640, 1280, 2560, 5120 ms. Even with all the above 3) Interference randomization for NPUSCH:
enhancements made to NB-IoT in 3GPP release 15, the It was identified that NPUSCH channel suffers significant
NPRACH receiver design proposed in this paper should be degradation when operating in interference limited scenarios
agnostic to any changes in the number of symbol group (see [24] and the references therein). Proposals to introduce
frequency hopping tones. Also, for RTD estimation, in case of interference randomization based on SC-FDMA symbol-level
𝐺 < 𝑃, the symbol groups may have to be non-coherently
scrambling for Rel-14 unicast NPUSCH configured with
combined for RTD estimation if the gap between symbol
interference randomization enhancement or Msg3 in non-
groups two contiguous symbol groups is larger than the channel
anchor carrier are currently being discussed. This doesn’t
coherence time. Note that non-coherent combining may be
impact the NPUSCH receiver design proposed in this paper
required for only TDD, since 𝐺 = 𝑃 for FDD. With 3-level
hopping defined for preamble Format 2, the parameter 𝑇𝑙 except that the new data unscrambling module must be
defined in (14) should be placed appropriately in 𝐹𝜌 . For introduced after demodulation, whenever it is defined in the
3GPP specification.
example, for preamble format 2, modify 𝑇𝑙 such that, 𝛼0 = 1,
𝛼1 = 3, and 𝛼2 = 18 and construct the vector 𝐹𝜌 appropriately
7. CONCLUSION
for detection.

Preamble 𝑮 𝑷 𝑵
𝑻CP 𝑻SEQ In this paper study of the receiver design for NB-IoT uplink
format channels was presented. Discussions were provided in detail for
66.67 𝜇𝑠 or
0 4 4 5
2048Ts 5  8192Ts each uplink channel for time, frequency and signal structures.
Receiver design guidelines were provided for all three channels
266.67 𝜇𝑠 or
1 4 4 5
8192Ts 5  8192Ts with detailed mathematical analysis. The performance of each
channel was investigated and compared against the 3GPP
800 𝜇𝑠 or
2 6 6 33∙ 24576𝑇s RAN4 requirements with both link level simulations and
24576Ts
implementation on a commercially deployed Qualcomm® FSMTM
Table 4: NPRACH Preamble formats for FDD in 3GPP Release 15
Small Cell platform. Various practical implementation issues
Preamble 𝑮 𝑷 𝑵 were considered and enhancements to NB-IoT in 3GPP Release
𝑻𝑪𝑷 𝑻SEQ 15 were discussed. It was shown how the NB-IoT uplink
format
0 2 4 1 150.5 𝜇𝑠 1 8192Ts receiver design proposed in this paper can be adapted to include
or the proposed NB-IoT enhancements in 3GPP Release 15 with
4778Ts minor or no modifications. With NB-IoT being adopted by
1 2 4 2 266.67 2  8192Ts various network operators across the world, the work in this
𝜇𝑠 or
paper is of significance to system designers looking to
8192Ts
implement efficient NB-IoT uplink receiver on practical
2 2 4 4 266.67 4  8192Ts systems to work in conjunction with legacy LTE processing.
𝜇𝑠 or
8192Ts
0-a 3 6 1 50 𝜇𝑠 or 1 8192Ts 8. ACKNOWLEDGEMENTS
1536Ts The author would like to thank Raja Bachu for his guidance
1-a 3 6 2 100 𝜇𝑠 2  8192Ts in system design aspects; firmware and test teams in Qualcomm
or India Private Limited (QIPL), Hyderabad, for all the efforts in
3072Ts developing and testing the proposed receiver algorithms.
Table 5: NPRACH preamble formats for TDD in 3GPP Release 15.
9. APPENDIX
2) NPUSCH Collision with NPRACH:
In RAN1#95, the following was agreed that for the new
NPRACH resources introduced in Release15 with early data

2327-4662 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2019.2957641, IEEE Internet of
Things Journal
13

𝑅𝑈
A. Definition of mathematical symbols used in derivations. 𝑁𝑆𝐶 Number of consecutive subcarriers in an UL
resource unit for NB-IoT
𝑁𝑃𝑈𝑆𝐶𝐻
Symbol Definition 𝑀𝑖𝑑𝑒𝑛𝑡𝑖𝑐𝑎𝑙 Number of repetitions of identical slots for
𝑡ℎ 𝑡ℎ
is the transmit signal at 𝑛 sample of 𝑖 NPUSCH.
𝑠𝑚,𝑖 [𝑛]
symbol in 𝑚𝑡ℎ symbol group 𝑃𝑘 phase difference across adjacent subcarriers 𝑘
Received time domain signal of 𝑛𝑡ℎ sample of 𝑄𝑘𝑠 phase difference across adjacent subcarriers 𝑘
𝑦𝑚,𝑖 [𝑛] in slot 𝑠.
𝑖 𝑡ℎ symbol in 𝑚𝑡ℎ symbol group
Channel gain of 𝑖 𝑡ℎ symbol in 𝑚𝑡ℎ symbol ℎ̂𝑠,3 Frequency smoothed channel estimate in slot 𝑠
ℎ𝑚,𝑖 for NPUSCH format 1.
group.
Received frequency domain signal of 𝑘 𝑡ℎ tone 𝐻𝑣 FFT vector of frequency smoothed channel
𝑌𝑚,𝑖 [𝑘] estimates.
of 𝑖 𝑡ℎ symbol in 𝑚𝑡ℎ symbol group
Frequency domain samples of 𝑖 𝑡ℎ symbol in 𝐼𝑣 Raw (not normalized) CFO detection metric.
𝑌𝑚,𝑖 𝑚𝑚𝑎𝑥 Maximum of CFO detection metric.
𝑚𝑡ℎ symbol group.
ℎ̂𝑏 Average channel estimates for block 𝑏.
Sum of frequency domain samples of 𝑚𝑡ℎ
𝑌𝑚 𝑒𝑠,𝑙,𝑘 Equalized symbol 𝑙, for slot 𝑠, and tone 𝑘.
symbol group. 𝑡 (𝑛)
Differential metric of 𝑚𝑡ℎ symbol group.
𝑒𝑠,𝑙,𝑘 Time domain sample 𝑛 of equalized symbol 𝑙,
𝑍𝑚 for slot 𝑠, and tone 𝑘.
Sum and difference of differential metrics 𝑙𝑙𝑟𝑠,𝑙,𝑘 Log-likelihood ratios for symbol 𝑙, for slot 𝑠,
𝑋𝑚1 & 𝑋𝑚2
across two symbol groups respectively. and tone 𝑘.
Sum and difference of differential metrics 𝑆𝑁𝑅𝑏 Signal to noise ratio (SNR) for block 𝑏.
𝑊1 & 𝑊2
across symbol groups and repetitions 𝑍𝑠,𝑙 [𝑘0 ] Phase de-rotated frequency domain samples
respectively. for slot 𝑠 and symbol 𝑙.
Final differential metric with phase p
𝑍𝑠 Combined phase de-rotated frequency domain
proportional to the estimated frequency offset. samples for pilot symbols in slot 𝑠.
𝑊 𝑍𝑠𝑑 Combined phase de-rotated frequency domain
Frequency offset corrected differential. samples for data symbols in slot 𝑠.
𝑉𝑚
𝐽𝑟 (𝑢) Equalized, frequency offset corrected and
Frequency offset corrected and frequency
𝑈4𝑟+𝑙 accumulated pilot and data symbols over RUs
hopping corrected (conjugated) differential.
and repetitions.
Frequency offset corrected and frequency 𝐽(𝑢)
𝑇𝑙 Detection metric for NPUSCH Format 2
hopping corrected (conjugated) differential
received symbols.
summed across repetitions.
𝑠(𝑡) NB-IoT time domain sample at time 𝑡.
FFT vector of corrected differentials. 𝑘 NB-IoT tone index.
𝑄𝑉
Raw (not normalized) NPRACH detection 𝑎𝑘 Modulated alphabet transmitted on tone 𝑘.
𝐺𝑣 Δ𝑓 Subcarrier spacing between NB-IoT tones.
metric vector.
𝑁𝑐𝑝,𝑙 Number of cyclic prefix samples of symbol 𝑙.
Maximum value of the raw (not normalized)
𝐺∗ 𝑇𝑠 Sampling time interval (Basic Time Unit in [10]).
NPRACH detection metric vector after
𝑓𝑜 Frequency separation between centers of LTE
polynomial interpolation.
and NB-IoT carriers
SNR on the NPRACH subcarrier 𝜌 . 𝑀0 Time domain samples between beginning of
𝜆
𝑁𝑃𝑈𝑆𝐶𝐻 frequency shift application and the 𝑙 𝑡ℎ OFDM
𝑁𝑟𝑒𝑝 Scheduled number of repetitions of a NPUSCH
symbol.
transmission.
𝑈𝐿
𝑁𝑠𝑦𝑚𝑏 Number of assigned tones and number of
symbols in a slot respectively
𝑈𝐿
𝑁𝑠𝑙𝑜𝑡𝑠 Number of consecutive slots in an UL resource
unit for NB-IoT.
[4] R. Ratasuk, et.al, "Overview of narrowband IoT in LTE Rel-13,” IEEE
Conference on Standards for Communications and Networking (CSCN),
10. REFERENCES Berlin, 2016, pp. 1-7.
[5] X. Lin, A. Adhikary, and Y.-P. E. Wang, “Random Access Preamble
[1] “Small Cells taking cellular to new heights with IoT tech and global Design and Detection for 3GPP Narrowband IoT systems,” IEEE
deployments”, https://www.qualcomm.com/news/onq/2018/03/16/mwc- Wireless Communications Letters, vol. 5, no. 6, pp. 640–643, Dec. 2016.
2018-small-cells-taking-cellular-new-heights-iot-tech-and-global- [6] R. Harwahyu, R. G. Cheng, C. H. Wei and R. F. Sari, "Optimization of
deployments , Mobile World Congress (MWC), March 2018. Random Access Channel in NB-IoT," in IEEE Internet of Things Journal,
[2] 3 rd Generation Partnership Project, “NB-LTE – General L1 Concept vol. 5, no. 1, pp. 391-402, Feb. 2018.
Description”, Ericsson, et.al, R1-156010, Oct. 2015. [7] R. Ratasuk, et.al, "Data Channel Design and Performance for LTE
[3] Y.-P.Eric Wang et.al, “ A Primer on 3GPP Narrowband Internet of Narrowband IoT," IEEE 84th Vehicular Technology Conference (VTC-
Things (NB-IoT)”, IEEE Communication magazine Vol 55, issue 3, Fall), Montreal, QC, 2016, pp. 1-5.
March 2017.

2327-4662 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2019.2957641, IEEE Internet of
Things Journal
To be submitted to IEEE Journal. 14

[8] J. Hwang, C. Li and C. Ma, "Efficient Detection and Synchronization of

Superimposed NB-IoT NPRACH Preambles," in IEEE Internet of Things
Arvind Chakrapani (M’2012) received the
Journal. vol. 6, no. 1, pp. 1173-1182, Feb. 2019.
[9] 3 rd Generation Partnership Project, “NB-PRACH design- document for Ph.D. degree in electrical engineering from
discussion and decision”, R1-160316, Huawei, HiSilicon, Neul, 3GPP, the University of New South Wales, Sydney,
TSG RAN WG1 Meeting #84, Feb 2016. NSW, Australia, in 2011. He is currently a
[10] 3 rd Generation Partnership Project, “Technical Specification Group
system engineer with Qualcomm Flarion
Radio Access Network; Evolved Universal Terrestrial Radio
Access; Physical Channels and Modulation”, 3GPP TS 36.211 Technologies (part of Qualcomm
V14.1.0., Dec. 2016. Technologies Incorporated), Bridgewater,
[11] 3 rd Generation Partnership Project, “Technical Specification Group New Jersey, where he has been involved in
Radio Access Network; Evolved Universal Terrestrial Radio
the research in the areas of resource
Access; Physical layer Procedures”, 3GPP TS 36.213 V14.1.0.,
Dec. 2016. allocation in LTE/eMTC scheduler; MAC
[12] 3 rd Generation Partnership Project, “Technical Specification Group and physical layer algorithms for eMTC and NB-IoT; and
Radio Access Network; Evolved Universal Terrestrial Radio physical layer algorithms for V2V technology. His current
Access; Radio Resource Control (RRC), Protocol specification”,
research interests include developing efficient physical layer
3GPP TS 36.331 V14.2.0., Mar. 2017.
[13] 3 rd Generation Partnership Project, “Technical Specification Group algorithms for initial access, control and data channels in 5G-
Radio Access Network; Evolved Universal Terrestrial Radio NR FR1/FR2.
Access; Medium Access Control (MAC) protocol specification”,
3GPP TS 36.321 V14.1.0., Mar. 2016.
[14] 3 rd Generation Partnership Project, “Technical Specification Group
Radio Access Network; Evolved Universal Terrestrial Radio
Access; Base Station (BS) radio transmission and reception”, 3GPP
TS 36.104 V14.5.0., Sep. 2017.
[15] 3 rd Generation Partnership Project, “Technical Specification Group
Radio Access Network; Evolved Universal Terrestrial Radio
Access; Base Station (BS) conformance testing”, 3GPP TS 36.141
V14.5.0., Sep. 2017.
[16] J.O. Smith, “Spectral Audio Signal Processing”, W3K Publishing,
http://books.w3k.org/, ISBN 978-0-9745607-3-1.
[17] 3 rd Generation Partnership Project, “Technical Specification Group
Radio Access Network; Evolved Universal Terrestrial Radio
Access; Radio Resource Control (RRC): Protocol Specification”,
3GPP TS 36.331 V14.2.0., March. 2017.
[18] R. Ratasuk, N. Mangalvedhe, Z. Xiong, M. Robert and D. Bhatoolaul,
"Enhancements of narrowband IoT in 3GPP Rel-14 and Rel-15," IEEE
Conference on Standards for Communications and Networking (CSCN),
Helsinki, 2017, pp. 60-65.
[19] A. Chakrapani, "Efficient resource scheduling for eMTC/NB-IoT
communications in LTE Rel. 13," IEEE Conference on Standards
for Communications and Networking (CSCN), Helsinki, 2017, pp.
66-71.
[20] Keysight Signal generator , MXG N5182B, web link
https://literature.cdn.keysight.com/litweb/pdf/5990-9959EN.pdf.
[21] Vertex Channel emulator, VR5, web link, https://www.spirent.com/-
/media/Datasheets/Mobile/VR5_Datasheet.pdf
[22] 3 rd Generation Partnership Project, ‘Summary of simulation results for
NB-IoT BS demodulation requirement’, R4-166484, 3GPP TSG-RAN
WG4 meeting #80, Aug. 2016.
[23] 3 rd Generation Partnership Project,” Feature lead summary
NPRACH reliability and range enhancements”, R1-1807429, 3GPP
TSG-RAN WG1 Meeting #93, May 2018.
[24] R. Ratasuk, et.al, "Enhancements of narrowband IoT in 3GPP Rel-
14 and Rel-15", IEEE Conference on Standards for
Communications and Networking (CSCN), pp. 60-65, Sept 2017.
[25] 3 rd Generation Partnership Project, “Technical Specification Group
Radio Access Network; Evolved Universal Terrestrial Radio
Access; and Evolved Universal Terrestrial Radio Access Network,
Overall Description”, 3GPP TS 36.331 V14.2.0., Mar. 2017.

2327-4662 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.