SPECIAL SECTION ON GREEN COMMUNICATIONS AND NETWORKING FOR 5G WIRELESS

Received October 1, 2016, accepted November 2, 2016, date of publication November 29, 2016,
date of current version January 4, 2017.
Digital Object Identifier 10.1109/ACCESS.2016.2628820

Smart Channel Sounder for 5G IoT: From

Wireless Big Data to Active Communication
XUHONG CHEN1 , SHANYUN LIU1 , JIAXUN LU1 , PINGYI FAN1 , (Senior Member, IEEE),
AND KHALED BEN LETAIEF2 , (Fellow, IEEE)
1 Department of Electronic Engineering, Tsinghua University, Beijing 100084, China
2 Department of Electronic and Computer Engineering, The Hong Kong University of Science and Technology, Hong Kong
Corresponding author: P. Fan ([email protected])
This work was supported in part by the State Key Development Program of Basic Research of China under Grant 2012CB316100(2) and in
part by the European Union Project CoNHealth under Grant 294923.

ABSTRACT Internet-of-Things (IoT) will connect billions of smart devices and generate inundant data
through prominent solutions, such as machine type communication. The Third Generation Partnership
Project has launched the corresponding standards for multiple heterogeneous wireless smart devices in the
long term evolution (LTE)/LTE-advanced. In the forthcoming years, the valuable information hidden in the
deluge of data will be extracted and utilized in every field to improve quality and efficiency. However, the
bottleneck of realizing this magnificent vista of future intelligent lives lies in how to satisfy the practical
demands to transmit huge data volume through efficient wireless communication in diverse scenarios.
Herein, multi-scenario wireless communication triggers critical problems in wireless channel modeling and
soundings for 5G IoT, which by far, are understudied. In this paper, we introduce a general wireless channel
model and its multiple up-to-date corresponding channel sounding methods for future 5G IoT green wireless
communication. Through adopting the perspective of wireless big data excavation, the smart channel sounder
transforms the traditional passive wireless communication scheme into an active expectation-guaranteed
wireless communication scheme, which helps achieve efficient and green communication. To demonstrate
the validity and efficiency of this smart sounder scheme, we make a compatible prototype testified in
multiple scenarios. The multiple real-scenario experiments demonstrate that the smart sounder can function
effectively, especially in those scenarios where traditional channel state information is not available or
imperfect.

INDEX TERMS 5G IoT, channel sounding, wireless big data, active estimation, green communication.

I. INTRODUCTION and facilitate future personal life, which can be applied in

Inundant data is being continuously generated not only by city-wide facilities and institutions like smart grid, manufac-
traditional personal computers, tablets and smart phones, but ture industry, medical industry, transportation, etc. However,
also by ever-growing smart devices like sensors, decision- realizing such a magnificent vista of future intelligent life
making devices and other miscellaneous electronic admea- requires revolution on diverse advanced technologies such as
suring apparatuses that are indiscriminately connected to communication, data mining, etc. In this paper, we focus on
the internet, which triggers the advent of the Internet-of- the new channel sounding method for future 5G IoT since
Things (IoT) era. According to the authoritative prediction efficient and stable wireless communication is the bottleneck
conducted by Gartner Inc. [1], more than 26 billion smart for volumes of data transmission, where we define 5G IoT
units will be connected by the year of 2020 as part of the as the internet-connected devices utilizing next generation
IoT. Each IoT device is deemed as a basic function unit of communication techniques to transmit data. It is estimated
an integrated cyber-physical ecosystem, which is capable to that by 2020, the total data generated by IoT devices will
independently or cooperatively execute multiple tasks after exceed 4.4 ZB (zettabytes) [3].
the valuable information is exploited out from the inundant One of the promising solutions for Machine-to-
raw data generated by massive IoT devices [2]. Therefore, IoT Machine (M2M) communication is Machine Type Commu-
is a promising integration system to nurture economy growth nication (MTC) which is capable to connect diverse IoT

2169-3536
2016 IEEE. Translations and content mining are permitted for academic research only.
8888 Personal use is also permitted, but republication/redistribution requires IEEE permission. VOLUME 4, 2016
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
X. Chen et al.: Smart Channel Sounder for 5G IoT

devices. Unlike traditional Person-to-Person (P2P) communi- detection may not be indispensable since it will consume
cation scheme, the design and application of M2M communi- more energy due to frequent channel detections and the
cation like MTC encounters new challenges in IoT scenario. estimation results may not be perfect due to the low SNR
The IoT devices may be deployed in wide geographical actuality.
ranges to certain places that people rarely get to where the However, for certain IoT devices, they are installed in fixed
communication scenarios varies a lot. Therefore, traditional locations once they are deployed. If the statistical channel
channel modeling methods may not cover all the possible sce- fading information (CSI) in those scenarios can be acquired
narios. For example, sensor networks can be installed widely through estimating the m value of Nakagami-m fading, the
in the open air, the wireless channel under this situation may channel quality can be forecasted and evaluated according
not be traditional Rician channel or Rayleigh channel. to the obtained statistical m value before the data transmis-
Aiming to address the aforementioned problems, we sion. Since the ready-to-transmit data is delay-tolerant, it is
introduce a general channel model Nakagami-m fading favourable to transmit the same amount of data with lower
channel. Up to now, the Nakagami-m fading model is a power consuming when the channel is in good condition
prominent candidate for future broadband green wireless (i.e relatively large m value and relatively high SNR). There-
communication system for its superior performance since it is fore, the communication scheme changes from conventional
more versatile in characterizing various fading channels than blind communication, where the CSI is not know before
Rayleigh and Rician fading [5], [6] according to its definition. transmission but through frequent and instant channel detec-
It is well known that the probability density function (pdf) of tions, to sorts of smart green communication, where the CSI
Nakagami-m distribution is [7]: can be predicted through analyzing the statistical data [8].
2 m m 2m1 mr 2 /  Consequently, to realize such a smart green com-
PR (r) = ( ) r e ,r 0 (1) munication scheme lies in how to efficiently obtain the
0(m) 
accurate statistical m value in IoT scenarios. Accord-
and its k-th moment is ing to the literature, accurate m parameter estimation of
0(m + k/2)  k/2 Nakagami-m fading channel under certain scenarios has been
k = E[Rk ] = ( ) (2)
0(m) m studied [9][14]. It is proved that the acquirement of sta-
thus, tistical m parameter is crucial for a large range of possible
potential industrial designs and service optimizations of IoT
2 1
m= , m (3) devices [15][17], especially when considering the effect of
E[(R ) ]
2 2 2 fading index m value on the wireless channel capacity estima-
where R is a random variable obeying Nakagami-m distri- tion [18]. Therefore, if the statistical properties of m value of
bution and 0() represents the standard Gamma function. certain scenarios of interests are pre-studied as [19], it would
 = 2 = E[R2 ] is the second moment. In particular, be better to design wireless transmission schemes for diverse
for m = 0.5, the Nakagami-m fading is identical to one- content-aware transmissions [20] by taking advantage of the
sided Gaussian fading. For m = 1, the Nakagami-m fading is m value in advance, especially for those critical scenarios
degraded into Rayleigh fading. For m = (K + 1)2 /(2K + 1), where conventional approaches [21] for instantaneous CSI
the Nakagami-m fading approximates Rician fading with a are not always available or imperfect [22]. In brief, the m
parameter K . For the case m = , it stands for no fading. value is an indicator of the channel fading degree and is
Apart from channel modeling, the biggest challenge comes valuable for the appropriate customization and deployment
from the enormous data requiring transmission service from of IoT devices.
huge number of different IoT devices. Dedicated communi- Previous works [23][28] on the estimation of m parameter
cation system for P2P communication that supports high-data mainly focused on developing estimation algorithms in high
rate and latency-intolerant service may not entirely suitable SNR regime by ignoring the system implementation com-
for MTC since the massive data transmission is delay-tolerant plexity, usually, this leads to superior in-lab testing perfor-
and high-data rate is not always indispensable, which enlight- mance but potentially inferior real-time real-scenario estima-
ens us to redesign the channel sounding scheme according to tion
the different transmission requirements. performance. Moreover, in-lab software simulation can nei-
Energy-saving is another crucial issue for MTC because ther acquire real m value nor verify the validity of the esti-
the build-in battery of wide-deployed IoT devices is con- mation algorithms. Up to now, few works have successfully
strained. It is true that due to the limited energy, most of implemented and tested the off-the-shelf m parameter esti-
the IoT is in transmission-sleep mode for most time and the mation algorithms in real-scenario estimations to prove their
data amount per transmission is relatively small that it is validity and efficiency [29].
not efficient to detect the channel for small size of data per In this paper, we introduce Nakagami-m fading model into
transmission. Besides, the installation places of IoT devices 5G IoT scenarios, based on which we endeavour to upgrade
may be like underground parking sites where the penetrate the conventional blind communication into expectation-
loss is severe, which leads to low signal-to-noise ratio (SNR) guaranteed smart green communication through mining the
due to limited transmission power. Therefore, instant channel statistical m value for channel fading information. Through

VOLUME 4, 2016 8889

 X. Chen et al.: Smart Channel Sounder for 5G IoT

analyzing diverse estimation methods, we find out that our indicated that the biases of this algorithm are relatively small
proposed estimation algorithm for m value under low SNR and close to the Cramer-Rao lower bound. However, most
regime can function well in 5G IoT scenarios. To further ver- ML-based algorithms had an underdeveloped performance
ify the validity and feasibility of this smart sounding method, in low SNR regime. To address this issue, an estimation
we made a prototype of this smart channel sounder, which is algorithm combining ML-based and moment-based method
mainly implemented on FPGA associated with a graphic user was proposed by Liu et al. [30]. By taking the noise infor-
interface (GUI) design. Besides, a general hardware frame- mation into account, the proposed algorithm has much better
work for a quick verification and application of the validity of performance than previous known algorithms especially in
other newly invented estimation algorithms for m parameter low SNR regime.
are presented as well, which makes both the acquirement For other uncommon algorithms, Yip and Ng [24] put
of real m parameter and the performance upgradation of forward a simulation model for the Nakagami-m fading under
related algorithms in 5G IoT scenario sounding feasible. The the condition m < 1, which is conducted by employing a
hardware prototype owns an up-to-date estimation accuracy square-root-beta process and a complex Gaussian process.
with flexible scalability and wide working frequency band Chai and Alouini [28] presented a software-based m parame-
ranging from 400MHz to 6GHz. The reasonable complexity ter estimator with potential applications of m parameter that
level with adjustable estimation accuracy makes the total esti- aimed at optimizing the power allocation in the case when the
mation time short enough so that this prototype is applicable instant CSI estimation was imperfect.
to 5G IoT scenarios to obtain the statistical channel state It is worth mentioning that few works had contributed to
information where conventional approaches are incapable. the prototype-making of related estimation and variate gen-
The acquired statistical m data can be directly utilized in eration of Nakagami-m fading, due to complexity of hardware
transmission scheme designs, optimal location distribution of design. Alimohammad et al. [31] put forward an FPGA-based
IoT devices and optimal transmission power allocation. hybrid hardware variate generator for both Nakagami-m and
The remainder of this paper is organized as follows. Weibull distributions. Maletic et al. [32] proposed a hardware
Section II introduces the related diverse estimation methods assessment method of impairments concerning Nakagami-m
on the m parameter, the related works on smart sounding fading. However, the prototype-making of the m parameter
prototype. Section III presents prototype details containing estimation was not considered especially for 5G IoT scenario.
the detailed FPGA implementation and the general hard-
ware framework design. In Section IV, the multi-scenario III. THE SMART CHANNEL SOUNDING PROTOTYPE
estimation results and hardware framework evaluation are This section introduces the detailed implementation process
presented. In Section V, we give out two potential application and the hardware framework design of the smart channel
scenarios where this smart channel sounder can be applied. sounding prototype. The implemented estimation algorithm
Finally, Section VI concludes the paper with a brief outlook is mainly based on Liu et al. [30] compared with other clas-
for future work. sical methods because this algorithm outperforms any other
existing algorithms in low SNR regime. The main object of
II. RELATED WORKS this prototype is to implement a hardware platform that can
Common m parameter estimation algorithms over software- both actually measure the m value in real-scenario estimations
based simulator falls into two categories, namely the and verify the newly-proposed m parameter estimation algo-
moment-based algorithms and the maximum-likelihood (ML) rithms quickly. Therefore, the proposed prototype benefits the
based algorithms, each possessing its own merits and draw- design and application of IoT devices in terms of installation,
backs. operation, and other side information.
For moment-based algorithms, Cheng and Beaulieu [10]
proposed the first moment-based method to estimate the m A. TRANSMISSION SYSTEM MODEL
parameter of Nakagami-m fading in 2001. To obtain pre- The wireless transmission baseband model can be expressed
cise results, the simulation requires a large number of sam- as:
ples. Besides, the estimation fluctuates intensely in the low
SNR regime [11]. Tepedelenlioglu and Gao [13] presented a yi = hi ri xi + ni , i = 1, 2, ldots, N (4)
novel integer moment-based estimator which considered both
the estimation adaptation and the computational complex- where hi and ri denote the large-scale attenuation and small-
ity. A more generalized method of moment-based (GMM) scale fading channel gains, respectively. ni represents the
algorithm introduced by Wang et al. [14] achieved nearly independent and identically distributed additive white Gaus-
identical asymptotic performance with ML-based estimation. sian noise which obeys N (0, 2 ), 2 > 0. xi is the length-N
However, this algorithm is very complicated and therefore is training symbol and ri is a Nakagami-m distributed variable
unlikely to be applied to real-scenario estimations. which is the main concern in this paper.
For ML-based algorithms, Cheng and Beaulieu [12] firstly In the real-scenario estimation of 5G IoT scenarios, the
proposed an ML-based estimation algorithm for m parameter channel gain hi usually remains unchanged during channel
estimation of Nakagami-m fading. The simulation results coherent-time, even in high-mobility scenarios. Thus, both

8890 VOLUME 4, 2016

X. Chen et al.: Smart Channel Sounder for 5G IoT

to the external memory. Besides, the GUI module in Fig. 2

serves as a control terminal and a dynamic output display
platform. The estimation results are displayed in terms of both
curve graphs and statistical tables, which will be illustrated
later.

C. ESTIMATION PROCESS FOR m PARAMETER

Most of conventional approaches for m parameter estimation
ignore the noise factor whereas the algorithm implemented
in this paper considers it as a valuable side information.
Therefore, before the estimation process for m parameter,
FIGURE 1. Block diagram of the m estimation algorithm. the noise power spectrum of the environment needs to be
estimated. The promising methods can either be what is
described as in Gholizadeh et al. [27] or just through manu-
sides of Eq. (4) can be normalized through dividing hi xi : ally dysfunctioning the transmission parts for several frames.
After careful comparison of the estimation performance of
yi = ri + ni , i = 1, 2, . . . , N (5) the two methods, the later one is chosen since it will not
where yi denotes the normalized received signal samples and introduce extra systematic complexity and meanwhile main-
tain the same performance within a short period of estimation
ni CN (0, 2 2 ) is the normalized noise samples, which
2
hi xi time which is crucial in 5G IoT scenarios estimation for m
obeys the Gaussian distribution. parameter.
The estimation process for one-time estimation is illus-
B. FPGA IMPLEMENTATION OF m PARAMETER trated in Fig. 3. The first step is to evaluate the environment
ESTIMATION noise and the second step is to estimate the m value. Both
Fig. 1 illustrates the block diagram of the implemented m of the two processes are performed under the supervision
parameter estimation algorithm on FPGA. The overall archi- of the GUI terminal. Since the channel gain can be fast
tecture consists of three parts: signal generating and sample time-varying and extremely unstable for disparate scenarios,
receiving module, correction factor calculation module and m for the sake of estimation accuracy and meanwhile consid-
parameter estimation module. ering estimation duration, the noise power density is esti-
The signal source xi is a pre-known signal with fixed mated for multiple times in this prototype to average out the
pattern which plays a similar role as the pilot used in the fluctuation.
CSI estimations. In particular, this pattern can be altered to
adjust the estimation accuracy according to diverse scenario D. GENERAL HARDWARE FRAMEWORK DESIGN
requirements. In consideration of enabling this prototype to easily update
In the correction factor module, it mainly calculates the the improved estimation algorithms while in the future con-
correction factor , which is extracted by taking advantage of veniently comparing the performance of diverse algorithms
the noise information. To be specific, the first up to the fourth on the same prototype, it is favorable to design a more
moments of yi are calculated for later calculation. Afterwards, general hardware framework for m parameter estimation of
the eigenvalue equation is solved using the dichotomy method Nakagami-m fading channel. To the best knowledge of the
to acquire the correction factor in the labeled DSE part. authors, this is the first general m parameter estimator proto-
Finally, the correction factor is generated with the help of type that can be directly utilized by other researchers studying
signal source xi . the same topic. With this general hardware design, the veri-
The estimator module is the final stage to calculate the fication process of any newly invented estimation algorithms
estimated m value with stored information such as the signal is much more convenient and also is more compatible with
source xi , the correction factor and the normalized received real m value in real-scenario estimation.
sample yi . The corresponding estimation algorithm can be The main parts of top-level block diagrams of the hardware
expressed as: framework design are illustrated in Fig. 2, which are mainly
p
6 + 36 + 481(y) composed of four parts: the FPGA module, the hardware
m= (6) board module, and the general interface and control terminal
241(y)
with GUI. The FPGA part is mainly responsible for generat-
N N
1 X 2 1 X ing training symbols and other indispensable transformations
1(y) = ln yi ln yi 2 (7)
N N of baseband signals. The hardware part is mainly responsible
i=1 i=1
for indispensable signal transformations between radio fre-
where yi = yi + . quency and media frequency. The general interface possesses
Under the supervision of memory controller, all the afore- four main functions, namely, the estimation parameter control
mentioned three modules of the estimation can freely access for m parameter estimation process, the input/output (I/O)

VOLUME 4, 2016 8891

 X. Chen et al.: Smart Channel Sounder for 5G IoT

FIGURE 2. General hardware framework design.

TABLE 1. Frequency range options.

E. THE DESIGN FOR MULTIPLE-INPUT-MULTIPLE-OUTPUT

(MIMO) ORTHOGONAL-FREQUENCY-DIVISION-
MULTIPLEXING(OFDM)
FIGURE 3. One-time estimation process for m parameter.
SYSTEM
For those who need to conduct the estimation for the m param-
eter in MIMO-OFDM systems, the corresponding framework
data stream control, the FPGA initialization configuration and interface design are illustrated in Fig. 2 as well. This
control and the hardware initialization and estimation config- is a demo for a 22 MIMO-OFDM system which can be
uration control. The aforementioned interface is integrated in readily expanded to multiple antennas. The channel selection
the GUI which controls the whole estimation process. More- module plays a key role of channel selector where only one
over, the match filter and down sampler parts are designed antenna is required in the real-scenario estimation, which
for the consideration of various data types required by other is also controlled by GUI. One practical problem is that
parameter estimation algorithms. the filter in hardware part cannot recognize signals with 15
Other developers of m parameter estimation can easily KHz interval. However, the validity of this design can be
replace the algorithm part with software-based algorithm proved through independent parallel estimations over differ-
in programming language to verify the validity of newly ent frequency bands such as 2.4GHz band and 2.5GHz band
invented algorithms rather than converting them into FPGA in Table I.
design which is time-consuming. After the performance is
positively confirmed, the developer can complete the design F. GUI DESIGN
with hardware realization and replace the software-based The designed GUI interface is illustrated in Fig. 4. The GUI
algorithm with FPGA design to accelerate the estimation can display the estimation results dynamically, which is very
speed according to the diverse scenarios. convenient for estimators to observe the performance of the

8892 VOLUME 4, 2016

X. Chen et al.: Smart Channel Sounder for 5G IoT

density and the setting for training sequence length can be

accordingly adjusted.

TABLE 2. Complexity comparison.

TABLE 3. Hardware framework parameter.

FIGURE 4. Graphic user interface for m parameter estimation.

implemented algorithm synchronously, especially in real-

scenario testing. The GUI helps to draw a brief conclusion on
the fading degree. For conventional approach, however, plots
can only be obtained when all the estimation procedures have
been completed.
The estimation configuration parameters like working fre-
quency range and training signal length can be initialized
through this GUI. Meanwhile, the GUI allows the operator IV. MULTI-SCENARIO ESTIMATION RESULTS
to start/pause the estimation process and to load/save the The algorithm for m parameter estimation has been imple-
estimation results in several forms. mented with VHDL and Verilog hardware description lan-
guage. Besides, the hardware framework has been realized
G. RECONFIGURATION AND COMPLEXITY with certain scripting languages. The FPGA is equipped with
The working frequency range options are presented in two DDR3-800 EDD RDIMM and the resource consumption
TABLE I. Diverse filter combinations will be enabled is comparatively low. Since future updating is also consid-
according to the configuration on the GUI. The constellation ered, the implementation is flexible for potential improve-
mapper possesses the choices of Quadrature Phase Shift Key- ments and the initialization parameter will be configured
ing (QPSK), 16 Quadrature Amplitude Modulation (QAM) through GUI. Therefore, this framework is a state-of-the-art
and 64QAM. Other detailed reconfigurations such as ampli- m parameter estimation prototype.
fier settings and training sequence length settings are omitted. The detailed adjustable parameters of this prototype are
Except for the aforementioned performance reason, the given by TABLE III. These parameters can be altered through
original motivation to implement this m parameter esti- the designed GUI according to specific requirements from
mation algorithm rather than other algorithms lies in the the user or the scenario. It should be emphasized that if the
tradeoff between comparative low computational complex- estimated m value is less than 0.5, the estimation is invalid by
ity and system performance. The computational complexity the definition of the m parameter.
of this implemented algorithm to measure correction fac-
tor is O(N log M ), where N stands for the training symbol A. SOFTWARE VS. HARDWARE ESTIMATION RESULTS
length and M represents a parameter concerning equation Due to the fact that this is the first and state-of-the-art pro-
solving accuracy with dichotomy. The residual computa- totype for m parameter estimation of Nakagami-m fading,
tional complexity that comes from m parameter estimation it is inevitably necessary to evaluate the performance of the
is O(N ). Therefore, the total computational complexity is proposed estimation method by comparing the hardware esti-
O(N log M ) + O(N ) which is equal to O(N log M ). Other mation results with software simulations.
methods to diminish systematic computational complexity Fig. 5 shows a comparative estimation results of soft-
includes replacing floating point arithmetic with fixed point ware simulation and hardware estimation. The hardware esti-
arithmetic. Moreover, the estimation time for noise power mation exhibits reasonable fluctuation which reflects the

VOLUME 4, 2016 8893

 X. Chen et al.: Smart Channel Sounder for 5G IoT

FIGURE 5. Theory-estimation comparison.

FIGURE 8. High-SNR estimation in 2.4GHz band.

FIGURE 6. Theory-estimation probability density comparison.

FIGURE 9. Low-SNR estimation in 2.4GHz band.

B. HIGH-SNR VS. LOW-SNR ESTIMATION RESULTS

The estimation results in different SNR regimes are respec-
tively illustrated in Fig. 8 and Fig. 9 to verify the proposed
algorithm in different conditions. The estimations are con-
ducted in 2.4GHz band for 50 times in both of the two cases,
where low SNR regime stands for SNR<0dB and high SNR
regime stands for SNR>5dB.
In the high SNR regime, the estimation results for two
FIGURE 7. Histogram comparison. (a) Real estimation. (b) Nakagami-m. moment-based algorithms [10], [13] fluctuate a bit whereas
the implemented algorithm [30] performs more stably.
The ML-based algorithm [12] possesses nearly the same
performance with that in [30]. Because the implemented
constant variation of the real channel state. Fig. 6 depicts the algorithm [30] extracts additional channel information as a
probability density distribution of Nakagami-m fading and compensation from the received noise information, the esti-
the real-scenario estimated fading, which indicates a high mation result is a little larger than that of [12], which is more
fitness between hardware estimation and software simulation. obvious in low SNR case. However, for two moment-based
Fig. 7 exhibits the statistical histogram comparison of algorithms, the results in low SNR regime are unsatisfactory
Nakagami-m fading and one-time real scenario estimation while the results of the two algorithms are acceptable. Several
results. The trend is similar but with slice disparities which invalid estimations may occur which make some algorithms
is caused by estimation time and training sequence length. not work well in the low-SNR regime.

8894 VOLUME 4, 2016

X. Chen et al.: Smart Channel Sounder for 5G IoT

FIGURE 10. Indoor estimation in 2.4GHz band. FIGURE 12. 2.4 GHz real-scenario estimation.

FIGURE 11. Outdoor estimation in 2.4GHz band.

FIGURE 13. 5 GHz real-scenario estimation.

C. IN-DOOR VS. OUT-DOOR ESTIMATION RESULTS is to conduct a potential trial for MIMO-OFDM systems
To test the versatility of this prototype in diverse scenarios under filter restriction.
which may also be the two typical communication scenarios The estimation results indicate that under the same condi-
for IoT devices, another control experiment is conducted. The tions, the fading is much severer for the 5GHz band than that
antenna distance and other parameter configurations are the for the 2.4GHz band. This is consistent with the fact that the
same and the results are exhibited in Fig. 10 and Fig. 11, propagation distance for high frequency wave is much shorter
respectively. In the indoor estimation, due to multiple barriers than low frequency waves under the same transmit power
and interference from other wireless devices working on the constraint. Therefore, the validity of the estimation results is
same frequency band, the estimated average value of m is verified to some extent.
smaller than that of the outdoor estimation. The implemented As to the MIMO-OFDM systems, the parallel estimation
algorithm still outperforms the two moment-based algorithms over different frequency bands can guarantee an acceptable
which have a little fluctuations. estimation performance in each sub-channel.
In the outdoor estimation, the four algorithms have a much
more similar performance and reasonable fluctuations. The E. TWO-DIMENSIONAL HEAT MAP OF m PARAMETER
average value of m is around 2.6, which indicates a medium As a simple example of the potential applications of this
fading severity. implementation and hardware framework in the future, mul-
tiple real-scenario estimations are conducted to acquire these
D. DIVERSE FREQUENCY BANDS ESTIMATION RESULTS heat maps of m parameter as illustrated in Fig. 14.
This estimation is conducted for two purposes as illustrated Fig. 14(a) and Fig. 14(b) are conducted in diverse fre-
in Fig. 12 and Fig. 13. The first one is to verify the large quency bands in the morning, where the transmitter is placed
estimation range of diverse frequency bands. The second one at the center of the laboratory. To attain each map, 100 time

VOLUME 4, 2016 8895

 X. Chen et al.: Smart Channel Sounder for 5G IoT

TABLE 4. Estimation duration and variance.

FIGURE 15. GUI display of multiple estimations. (a) 2.4GHz band.

(b) 2.5GHz band.

and statistical variance of the implemented algorithm are

illustrated in TABLE 4. According to the estimation statistics,
the variance can be reduced to a certain level with a larger
TSL or multiple NETs whereas the ED is several times longer.
Moreover, it is suggested to adopt 1024 bit TSL in real-
FIGURE 14. Heat map for m parameter. (a) 5GHz band. (b) 2.4GHz band.
(c) 2.4GHz band. scenario estimations especially for fast time-varying fading
channel and the NETs sets according to the estimation envi-
ronment.
repeated estimations at diverse spots are conducted with equal
spatial distance. According to the acquired data, simple intu- G. GUI DISPLAY
itive conclusion can be drawn from the heat map indicates The estimation results over the designed GUI of different
that fading in section C is much better than that in three frequency bands are exhibited in Fig. 15. These two esti-
other sections due to multiple indoor barriers. For the lower mations are conducted with receiver moving horizontally. In
frequency band, the fading in section A, B, and D are much this case, the estimators can observe the variation of the m
better than those in higher frequency band. parameter and make appropriate configuration modifications
Another bigger heap map is illustrated in Fig. 14(c). Owing accordingly. For other cases, the GUI can dynamically export
to the restriction of limited antenna wire, each section is and display the estimation results as well.
independently estimated with the transmitter placed in the
center. With a quick glance of this map, one can tell the gen- H. IMPLEMENTATION AND FRAMEWORK DESIGN RESULT
eral fading degree of each section and make decisions about Fig. 16 illustrates the FPGA implementation which is inte-
the distribution of wireless device. This map is beneficial grated in the white box in Fig. 17. The FPGA implementation
in a much larger scenario, e.g. an m parameter heat map of has been designed with most common components e.g. DSP
one base station transmission range, or an m parameter heat so that it is convenient to immigrate this design to other FPGA
map along the route of high-speed railway, which can statis- devices.
tically estimate the wireless channel capacity and therefore This framework design is portable to conduct field esti-
distribute the base station accordingly. mations with two transmit channels and receive channels for
parallel estimations. Most importantly, other developers of m
F. COMPLEXITY AND ACCURACY EVALUATION parameter estimation algorithm can quickly and effectively
Through altering the training sequence length (TSL) and verify the validity of the proposed algorithm, making some
noise estimation times (NETs), the estimation duration (ED) modifications accordingly.

8896 VOLUME 4, 2016

X. Chen et al.: Smart Channel Sounder for 5G IoT

FIGURE 18. Wireless communication system for HSR.

and frequency-selective fading. According the real-scenario

FIGURE 16. FPGA implementation. estimation [33] that the channel coherence time is less than
1ms, which makes the real-time channel tracking extremely
hard. Besides, as illustrated in Fig. 18, the dramatic fluc-
tuation of the average received signal-to-noise ratio (SNR)
caused by the high mobility of the train largely degrades the
performance of the conventional channel estimation methods
especially near the edge area of the base station (BS) cover-
age. The aforementioned problems challenges the traditional
channel sounding method with multiple difficulties. Even if
the traditional method can function anyhow, it is not efficient
and economy to apply it considering the tradeoff between
complexity and performance.
However, If the statistical m value along the rail is pre-
cognised through utilizing this prototype as in Fig. 17, the sta-
tistical channel capacity information can be predicted. If the
statistical channel sounding results of certain area along the
rail is always bad, the IoT devices onboard can store the data
FIGURE 17. The smart channel sounder prototype. in buffer until the channel is predicted to be good or the data
can be transmitted before the train reaches those area. There-
fore, the smart channel sounder can ameliorate the drawbacks
V. APPLICATION CASES of traditional blind communication with channel qualify-
A. CASE ONE guaranteed smart green communication, which is beneficial
If IoT devices are deployed in sites similar to our lab, the first to the transmission scheme design accordingly [34][36] and
thing we can do is to place the IoT devices in the location with consequently save energy.
good statistical fading condition according to the acquired
m value as illustrated in Fig. 14. After the IoT devices are VI. CONCLUSION AND FUTURE WORK
installed, the devices can store the ready-to-transmit data in A smart channel sounder and its prototype for 5G IoT devices
buffer first and wait for the predicted good channel state has been presented, which can acquire the statistical m value
to transmit data. For example, the interference in our lab is of diverse scenarios. Through analyzing the historical channel
serious when it is daytime due to multiple user access in state information, conventional blind communication can be
the same frequency band. Therefore, it is a good choice to updated to smart transmission-efficiency guaranteed green
transmit data in nighttime which is also can be predicted communication, which can optimize the transmission scheme
through statistical sounding and analyzing the acquired m according to the m value and consequently save energy for
value. IoT devices. Diverse m parameter estimation algorithms have
been tested on the designed prototype in multiple-scenarios
B. CASE TWO to simulate the 5G IoT scenarios, which exhibits a good
As demonstrated, the installed algorithm functions well in performance in different real-scenario estimations. The rec-
low SNR regime which is suitable for the wide massive ommended algorithm possesses a superior performance in
deployment of IoT devices. Here, we present another poten- low SNR regime among all the existing methods through
tial application case of this smart channel sounder. In mobility additionally introducing noise information as a compensation
scenario, especially in high-speed railway scenario, the effi- in the estimation process. The up-to-date general hardware
cient acquisition of perfect or imperfect CSI is nearly impos- framework with corresponding GUI design and wide estima-
sible due to the complexity of the diverse scenarios that one tion frequency ranges makes this prototype versatile in veri-
train may encounters, where the channel is fast time-varying fying and updating other m parameter estimation algorithms

VOLUME 4, 2016 8897

 X. Chen et al.: Smart Channel Sounder for 5G IoT

quickly. The estimation duration is considered appropriately [18] R. Zouheir and M. S. Alouini, On the capacity of Nakagami-m fading
in this design with a tradeoff against estimation accuracy channels with full channel state information at low SNR, IEEE Wireless
Commun. Lett., vol. 1, no. 3, pp. 253256, Apr. 2012.
according to diverse estimation environments. [19] R. Lorenzo, R. Juan, and N. Cardona, Evaluation of Nakagami fading
Several potential applications of this prototype over dif- behavior based on measurements in urban scenarios, AEU-Int. J. Elec-
ferent scenarios have been proposed which can be beneficial tron. Commun., vol. 61, no. 2, pp. 135138, Feb. 2007.
[20] E. Danish, V. D. Silva, A. Fernando, C. D. Alwis, and A. Kondoz,
to the design, transmission strategy optimization and the Content-aware resource allocation in OFDM systems for energy-efficient
appropriate spatial distribution selection of IoT devices. video transmission, IEEE Trans. Consum. Electron., vol. 60, no. 3,
Future work includes two possible directions. The first pp. 320328, Aug. 2014.
[21] M. Hsieh and C. Wei, Channel estimation for OFDM systems based
direction is to keep developing the m parameter estimation on comb-type pilot arrangement in frequency selective fading channels,
algorithm and quickly verifying it on this designed prototype. IEEE Trans. Consum. Electron., vol. 44, no. 1, pp. 217225, Aug. 1998.
Second, since this prototype can be massively installed over [22] X. Chen, J. Lu, and P. Fan, General hardware framework of Nakagami
m parameter estimator for wireless fading channel, in Proc. IEEE Int.
the city, the acquired statistical m data can be city-wide, which Workshop High Mobility Wireless Commun., Oct. 2015, pp. 15.
can be achieved through customized smartphone applica- [23] T. M. Wu and S. Y. Tzeng, Sum-of-sinusoids-based simulator
tions. The acquired data should be well-studied and utilized to for Nakagami-m fading channels, in Proc. IEEE Veh. Technol.
Conf. (VTC-Fall), Oct. 2003, pp. 158162.
optimize the overall performance of both the 5G IoT systems [24] K. W. Yip and T. S. Ng, A simulation model for Nakagami-m fading
and IoT devices design in the perspective of wireless big data. channels, m<1, IEEE Trans. Commun., vol. 48, no. 2, pp. 214221,
Feb. 2000.
REFERENCES [25] Q. T. Zhang, A note on the estimation of Nakagami-m fading parameter,
IEEE Commun. Lett., vol. 6, no. 6, pp. 237238, Jun. 2002.
[1] Gartner Says the Internet of Things Installed Base Will Grow to 26
[26] K. Zhang, Z. Song, and Y. L. Guan, Simulation of Nakagami fading
Billion Units by 2020, accessed on Dec. 12, 2013. [Online]. Available:
channels with arbitrary cross-correlation and fading parameters, IEEE
http://www.gartner.com/newsroom/id/2636073
Trans. Wireless Commun., vol. 3, no. 5, pp. 14631468, Sep. 2004.
[2] A. Bassi et al., Enabling Things to Talk: Designing IoT Solutions With the
[27] M. H. Gholizadeh, H. Amindavar, and J. A. Ritcey, Analytic Nakagami
IoT Architectural Reference Model. Berlin, Germany: Springer, 2013.
fading parameter estimation in dependent noise channel using copula,
[3] Accessed on Sep. 30, 2016. [Online]. Available: http://www.zdnet.
EURASIP J. Adv. Signal Process., vol. 1, pp. 111, Jul. 2013.
com/topic/the-power-of-iot-and-big-data/
[28] Y.-C. Ko and M.-S. Alouini, Estimation of Nakagami-m fading channel
[4] H.-W. Chang, M. C. Tseng, S. Y. Chen, M. H. Cheng, and S. K. Wen,
parameters with application to optimized transmitter diversity systems,
Field trial results for integrated WiMAX and radio-over-fiber systems on
IEEE Trans. Wireless Commun., vol. 2, no. 2, pp. 250259, Mar. 2003.
high speed rail, in Proc. Int. Symp. Pers. Indoor Mobile Radio Commun.
[29] X. Chen, S. Liu, and P. Fan, Hardware implementation on m parameter
(PIMRC), Toronto, ON, Canada, Sep. 2011, pp. 21112115.
ML estimation of Nakagami-m fading channel, in Proc. IEEE Int. Conf.
[5] A. Goldsmith, Wireless Communications. Cambridge, U.K.: Wireless Commun. Mobile Comput. Conf., Aug. 2015, pp. 13951399.
Cambridge Univ. Press, 2005, pp. 7879. [30] S. Liu, P. Fan, K. Xiong, and S. Yi, Novel ML estimation of
[6] T. Aulin, Characteristics of a digital mobile radio channel, IEEE Trans. m parameter of the noisy Nakagami-m channel, in Proc. IEEE Int.
Veh. Technol., vol. 30, no. 2, pp. 4553, May 1981. Workshop High Mobility Wireless Commun., Beijing, China, Dec. 2014,
[7] M. Nakagami, The m-distributionA general formula of intensity dis- pp. 188193.
tribution of rapid fading, Statist. Methods Radio Wave Propag., 1960, [31] A. Alimohammad, S. F. Fard, and B. F. Cockburn, Hardware implemen-
pp. 36. tation of Rayleigh and Ricean variate generators, IEEE Trans. Very Large
[8] G. Ding et al., On the limits of predictability in real-world radio spectrum Scale Integrat. (VLSI) Syst., vol. 19, no. 8, pp. 14951499, Jun. 2011.
state dynamics: From entropy theory to 5G spectrum sharing, IEEE Com- [32] N. Maletic, M. Cabarkapa, N. Neskovic, and D. Budimir, Hard-
mun. Mag., vol. 53, no. 7, pp. 178183, Jul. 2015. ware impairments impact on fixed-gain AF relaying performance in
[9] J. C. S. S. Filho, M. D. Yacoub, and G. Fraidenraich, A simple accurate Nakagami-m fading, Electron. Lett., vol. 52, no. 2, pp. 121122,
method for generating autocorrelated Nakagami-m envelope sequences, Jan. 2016.
IEEE Commun. Lett., vol. 11, no. 3, pp. 231233, Mar. 2007. [33] L. Liu, Position-based modeling for wireless channel on high-speed
[10] J. Cheng and N. C. Beaulieu, Moment-based estimation of the railway under a viaduct at 2.35GHz, IEEE J. Sel. Areas Commun., vol. 30,
Nakagami-m fading parameter, in Proc. IEEE Pacific Rim Conf. Commun. no. 4, pp. 834845, May 2012.
Comput. Signal Process., Victoria, BC, Canada, Aug. 2001, pp. 361364. [34] Y. Wang, C. Yang, and B. Ai, Iterative companding transform and filtering
[11] Y. Chen and N. C. Beaulieu, Estimation of Ricean and Nakagami distribu- for reducing PAPR of OFDM signal, IEEE Trans. Consum. Electron.,
tion parameters using noisy samples, in Proc. IEEE Int. Conf. Commun., vol. 61, no. 2, pp. 144150, May 2015.
Paris, France, Jun. 2004, pp. 562566. [35] D. Zhang, P. Fan, and Z. Cao, Interference cancellation for OFDM
[12] J. Cheng and N. C. Beaulieu, Maximum-likelihood based estimation systems in presence of overlapped narrow band transmission system,
of the Nakagami-m parameter, IEEE Commun. Lett., vol. 5, no. 3, IEEE Trans. Consum. Electron., vol. 50, no. 1, pp. 108114, Feb. 2004.
pp. 101103, Mar. 2001. [36] F. Zhang, Y. Wang, and B. Ai, Variable step-size MLMS algorithm for
[13] C. Tepedelenlioglu and P. Gao, Estimators of the Nakagami-m parameter digital predistortion in wideband OFDM systems, IEEE Trans. Consum.
and performance analysis, IEEE Trans. Wireless Commun., vol. 4, no. 2, Electron., vol. 61, no. 1, pp. 1015, Feb. 2015.
pp. 519527, Mar. 2005.
[14] N. Wang, X. Song, and J. Cheng, Generalized method of moments
estimation of the Nakagami-m fading parameter, IEEE Trans. Wireless
Commun., vol. 11, no. 9, pp. 33163325, Jul. 2012.
[15] H. Jabbar, Y. S. Song, and T. T. Jeong, RF energy harvesting system and
XUHONG CHEN received the B.S. and M.S.
circuits for charging of mobile devices, IEEE Trans. Consum. Electron.,
degree with the Department of Communica-
vol. 56, no. 2, pp. 247253, Feb. 2010.
[16] R. Zhang, B. Ai, L. Yang, H. Song, and Z. Li, A precoding and detec- tion Engineering from Chongqing University,
tion scheme for OFDM based wireless communication system in high- Chongqing, China, in 2010 and 2013, respec-
speed environment, IEEE Trans. Consum. Electron., vol. 60, no. 4, tively. He is currently pursuing the Ph.D. degree
pp. 558566, Nov. 2014. in electronic engineering with Tsinghua Univer-
[17] J. Park, K. Kim, J. Koo, and J. Kim, A low complexity soft-output sity, Beijing, China. His research interests include
detection algorithm for 22 multiple-input multiple-output multiband- high mobility broadband wireless communica-
OFDM systems using dual carrier modulation, IEEE Trans. Consum. tions, channel estimation, cross-layer design, and
Electron., vol. 60, no. 3, pp. 311319, Aug. 2014. network information theory.

8898 VOLUME 4, 2016

X. Chen et al.: Smart Channel Sounder for 5G IoT

SHANYUN LIU received the B.S. degree from the KHALED BEN LETAIEF (S85M86SM97
Department of Electronic Engineering, Tsinghua F03) received the B.S. (Hons.), M.S., and Ph.D.
University, Beijing, China, in 2014. He is currently degrees from Purdue University at West Lafayette,
pursuing the masters degree with the Depart- West Lafayette, IN, USA, in 1984, 1986, and 1990,
ment of Electronic Engineering, Tsinghua Univer- respectively, all in electrical engineering. In 1985,
sity. His current research interests include wireless he was with the School of Electrical and Computer
communications in high speed railway, parameter Engineering, College of Engineering, Purdue Uni-
estimation in Nakagami Channel, and information versity at West Lafayette, where he taught courses
theory. in communications and electronics as a Graduate
Instructor.
From 1990 to 1993, he was a Faculty Member with the University of
Melbourne, Australia. Since 1993, he has been with The Hong Kong Uni-
versity of Science and Technology, where he is currently a Chair Professor
and the Head of the Electronic and Computer Engineering Department. He is
also the Director of the Hong Kong Telecom Institute of Information Tech-
nology. His current research interests include wireless and mobile networks,
broadband wireless access, cooperative communications, cognitive radio,
JIAXUN LU received the B.S. degree from OFDM, CDMA, and Beyond 3G systems. In these areas, he has authored
Tsinghua University, Beijing, China, in 2014. over 300 journal and conference papers and given invited keynote talks as
He is currently pursuing the Ph.D. degree in well as courses all over the world.
electronic engineering from Tsinghua University. Dr. Letaief served as a Consultant for different organizations. He is the
His research interests include high speed railway founding Editor-in-Chief of the IEEE Transactions on Wireless Communi-
wireless communications, with a focus on signal cations. He has served on the editorial board of other prestigious journals
precoding and processing. including the IEEE Journal on Selected Areas in CommunicationsWireless
Series as the Editor-in-Chief. He has been involved in organizing a number
of major international conferences and events. These include serving as the
Technical Program Chair of the 1998 IEEE Mini-Conference on Communi-
cations Theory, a Co-Chair of the 2001 IEEE Communications Theory Sym-
posium, a Co-Chair of the 2004 IEEE Wireless Communications, Networks
and Systems Symposium, and a Co-Technical Program Chair of the 2004
IEEE International Conference on Communications, Circuits and Systems.
He is the General Chair of the 2007 IEEE Wireless Communications and
Networking Conference, WCNC07, and the Technical Program Co-Chair
of the 2008 IEEE International Conference on Communication, ICC08.
PINGYI FAN (M03SM09) received the B.S.
He served as the Chair of the IEEE Technical Committee on Personal
degree from the Department of Mathematics,
Communications. He is an IEEE Distinguished Lecturer of the IEEE Com-
Hebei University, in 1985, the M.S. degree
munications Society, and an Elected Member of the IEEE Communications
from the Department of Mathematics, Nankai
Society Board of Governors.
University in 1990, and the Ph.D. degree from the
Department of Electronic Engineering, Tsinghua
University, Beijing, China, in 1994. From 1997
to 1998, he visited The Hong Kong University of
Science and Technology as a Research Associate.
From 1998 to 1999, he visited the University of
Delaware, USA, as a Research Fellow. In 2005, he visited NICT of Japan as a
Visiting Professor. From 2005 to 2011, he visited The Hong Kong University
of Science and Technology for many times as a Visiting Professor. In 2011,
he was a Visiting Professor with the Institute of Network Coding, Chinese
University of Hong Kong. He is currently a Professor of the Department of
EE, Tsinghua University.
His main research interests include 5G technology in wireless communi-
cations such as Massive MIMO, OFDMA, Network Coding, Network Infor-
mation Theory, and Cross Layer Design. He is an Overseas Member of the
IEICE. He has attended to organize many international conferences including
as a General Co-Chair of the IEEE VTS HMWC2014, a TPC co-Chair of the
IEEE International Conference on Wireless Communications, Networking
and Information Security (WCNIS 2010), and a TPC member of the IEEE
ICC, the Globecom, the WCNC, the VTC, and the Inforcom. He has served
as an Editor of the IEEE Transactions on Wireless Communications, the
Inderscience International Journal of Ad Hoc and Ubiquitous Computing,
and the Wiley Journal of Wireless Communication and Mobile Computing.
He is also a Reviewer of over 32 international journals including 18 IEEE
Journals and eight EURASIP Journals. He has received some academic
awards, including the IEEE Globecom 2014 Best Paper Award, the IEEE
WCNC08 Best Paper Award, the ACM IWCMC10 Best Paper Award, and
the IEEE ComSoc Excellent Editor Award for the IEEE Transactions on
Wireless Communications in 2009.

VOLUME 4, 2016 8899