Proceedings, APSIPA Annual Summit and Conference 2021 14-17 December 2021, Tokyo, Japan

An Adaptive Rank Selection Method in 3GPP 5G

NR Systems
Wei-Hung Chou, Wei-Chen Pao, Chun-Chia Tsai, Ting-Yu Yeh, and *Jen-Yi Pan
Industrial Technology Research Institute, Taiwan (R.O.C.)
*National Chung Cheng University, Taiwan (R.O.C.)
Email: {AlexChou, WCPao, cctsai60030, tingyuyeh}@itri.org.tw, *jypan@ccu.edu.tw

Abstract — A user equipment (UE) estimates channel state to evaluate the system performance, including the system
information (CSI) which includes the rank indicator (RI), the efficiency and the error rate.
precoding matrix indicator (PMI), and the channel quality The remainder of this paper is organized as follows. Section
indicator (CQI), and feedbacks the CSI to the base station (BS). II describes the process of rank selection for downlink
Rank Indicator is a number (indicator) that represents how well a
Multiple Input Multiple Output (MIMO) system communication
transmission. We propose our adaptive rank selection method in
works. Therefore, the RI estimation and selection is an important section III and evaluate the performance in section IV. Finally,
issue for system efficiency. In this paper, we propose an adaptively section IV gives concluding remarks.
adjusted RI selection method to accommodate the channel
condition. Simulation results demonstrate that the performance of II. RANK SELECTION
the proposed adaptive method satisfies the performance
requirements of the IMT-2020 evaluation. Furthermore, the
A UE feedbacks the estimated RI to a BS in the uplink
proposed adaptive method works well in a low-mobility indoor direction according to its receiving capability. The BS
scenario. determines the number of layers based on the feedback RI. The
Index Terms— Rank selection, IMT-2020 process of RI feedback is shown in Fig. 1. The steps are as
follows [9]:
I. INTRODUCTION Step1. BS periodically transmits Channel State Information-
As the demand for a new generation of mobile networks Reference Signals (CSI-RS) to UE for CSI measurement.
grows, the 3rd Generation Partnership Project (3GPP) develops Step2. UE measures the received CSI-RS to obtain channel
new radio access technology (RAT) for fifth-generation (5G) state information. RI is estimated according to the channel
mobile networks, namely 5G New Radio (NR). It is a global situation.
standard for the air interface of 5G networks [1]. 3GPP submits Step3. UE feedbacks the estimated information to BS,
the 5G NR systems specification [1, 2] to International including RI, PMI, and CQI.
Telecommunication Union (ITU) as a candidate 5G mobile Step4. BS transmits the Physical Downlink Share Channel
communication system, which is expressed as the IMT-2020 (PDSCH) by considering the feedback information.
system.
5G NR systems support MIMO technologies [4] to meet a UE BS
high data rate requirement and maximize system capacity. Non-precoded
MIMO channel capacity highly depends on the rank, and the CSI-RS

rank depends on the antenna correlation, represented by RI.

Maximum RI number is related to the number of antennas, 1. Derive channel rank RI
CSI feedback
2. Derive codebook PMI
including the Tx or Rx. Maximum RI means no correlation 3. Based RI, PMI derive CQI
among antennas and no interference to each other. In this case, Transmits PDSCH based on
the system shows an outstanding performance [5]. Therefore, PDSCH
RI, PMI, and CQI

the rank selection is an important issue for system efficiency.

However, interference is unavoidable in real communication
systems. A real-time CSI estimation is not practical, either.
Fig. 1. An example of rank selection
Some researches have been studied in [6]. Based on the
characteristic of the time-varying channel, this paper proposes a
For a UE, a heuristic method to estimate RI is to use an
rank selection method that considers the channel variation in the
time domain. An adaptive approach with low complexity is exhaustive search method to evaluate its performance for each
applied to select RI to accommodate the channel variation. RI. UE returns a RI that could achieve maximum efficiency.
This paper evaluates the performance of the proposed However, this exhaustive search method takes a long time to
adaptive RI selection method according to the 5G NR system complete a RI selection since all cases are evaluated to obtain
specifications submitted by 3GPP [1, 2] and compares it with the system performance. In addition, it takes a delay time for the
the IMT-2020 performance requirements defined by the ITU [3]. RI feedback to the BS. This causes the channel state at the
In this paper, WiSE System Level Simulator (SLS) [7, 8] is used transmission time to be different from the estimated time.
Therefore, the exhaustive search method requires a high
computational time and shows an inaccurate estimation due to

978-988-14768-9-0/21/$31.00 ©2021 APSIPA 1912 APSIPA-ASC 2021

Proceedings, APSIPA Annual Summit and Conference 2021 14-17 December 2021, Tokyo, Japan

the time-varying channel. the last transmission is less than a threshold:

- If the BLER is less than a threshold
III. PROPOSED ADAPTIVE METHOD Increase the RI value
In the process of determining the layers, the BS determines - Else
the number of layers to transmit based on the feedback RI. In Use the returned RI
order to determine the RI, the BS could use a formula of an Step5. The BS transmits the PDSCH by considering the
MMSE capacity [10] to calculate the capacity of a rank for each feedback information.
subband. The formula is defined as

 
-1

Cr,b  log2 Ir +SINRr HeqH (r,b)Heq(r,b) (1)

where Cr,b is a capacity of a rank for each subband; Ir is the
identity matrix; SINRr is the SINR of each rank; Heq(r,b) is the
equivalent channel of the channel with the precoder of each rank
of each subband. The equivalent channel is defined as
H eq ( r ,b )  H b Vb ,r , r = 1,...,max rank (2)
where Vb,r is the precoder derived by using SVD. Hb is a subband
channel as the following.
Hb  Ub Vb (3)
After calculating the capacity of each subband, the capacity
of each RI is a sum of the subband capacity as the following.
Cr  b Cr,b (4) Fig. 2. The process of the proposed adaptive method

where Cr is the capacity of each rank. Therefore, a RI with the IV. SIMULATION RESULTS
maximum capacity is determined as WiSE SLS [7, 8] is used to demonstrate the performance of
 Rˆ   arg max
r
Cr (5) the RI selection methods, including the exhaustive search
Although the computational complexity compared to the method and the proposed methods with and without an adaptive
selection. The indoor hotspot scenario [3] is selected for
exhaustive search method is reduced, the channel variation and
simulation. The simulation parameters of the scenario refer to
the error rate tolerance are still not considered in the RI
the IMT-2020 evaluation, as shown in Table 1. The BLER
determination. Therefore, this paper further proposes an
threshold is configured as 0.1. The shift of RI value is configured
adaptive rank selection method to estimate RI using these two as one level. The system simulation time is one second which is
considerations. one hundred frames. The performance comparison includes the
A rank selection enhanced with an adaptive adjustment is average spectrum efficiency and the 5th percentile spectrum
proposed by further considering the channel variation and the efficiency under the FDD system.
error rate tolerance. The main idea is to make adjustments based
on the previous transmission situation and the feedback RI. The
error rate tolerance applied in this paper is Block Error Rate Table 1 SIMULATION ASSUMPTION FOR INDOOR-eMBB
(BLER). For example, when the BLER is low, the channel Parameter Value
condition is stable, and the data transmission is almost Carrier frequency 4 GHz
successfully received and decoded. When the BS knows that the BS antenna height 3m
BLER in the previous transmission is very low or under a Total transmit
21 dBm for 10 MHz bandwidth
power per TRP
threshold, the BS could assume that the feedback RI may adjust MS power class 23 dBm
to a higher level to boost the system performance. Therefore, the Inter site distance 20 m
RI value could be shifted to increase system performances by TRP antenna (M, N, P, Mg, Ng; Mp, Np) =
sacrificing the error rate. Otherwise, the RI value is determined configuration (4,4,2,1,1;4,4) for 32 ports;
without a shift. The steps of the proposed adaptive method are MS antenna (M, N, P, Mg, Ng; Mp, Np) =
configuration (1,2,2,1,1;1,1)
as follows: Antenna element
Step1. BS periodically transmits CSI-RS to UE for CSI 5 dBi for BS; 0 dBi for MS
gain
measurement. MS speeds 100% indoor, 3 km/h
Step2. UE measures the received CSI-RS to obtain channel Noise figure 5 dB for BS; 7 dB for MS
Traffic model Full buffer
state information. RI is estimated according to the channel 10 MSs per TRxP, randomly and
situation. MS density uniformly dropped throughout the
Step3. UE feedbacks the estimated information to BS, geographical area
including RI, PMI, and CQI. Channel model InH_A
Step4. BS further determines whether or not the BLER of

1913
Proceedings, APSIPA Annual Summit and Conference 2021 14-17 December 2021, Tokyo, Japan

A. Spectrum Efficiency Table 2 Simulation result comparison

Fig. 3 and Fig. 4 respectively represent the average spectrum 5th percentile
Average Spectrum
efficiency and the 5th percentile spectrum efficiency when the bits/sec/Hz Spectrum
efficiency
efficiency
UE speed is 3km/h. The green line in the figure represents the
Proposed adaptive
requirements of ITU-R for the indoor hotspot scenario of IMT- method
10.1914 0.345
2020 evaluation, which are 9 and 0.3 bits per second per Hz for OPPO [11] 10.410 0.300
the average spectrum efficiency and the 5th percentile spectrum Motorola Mobility
efficiency [3]. The red lines are simulation results of each of the 9.180 0.400
/ Lenovo [11]
three methods.

Fig. 5 represents the modulation coding scheme (MCS)

distribution. Interestingly, the MCS of the proposed method
without adaptive selection has 95% of MCS 27. MCS 27 is
represented as the code rate and the modulation order to the
highest level, shown in Table 3 [12]. The error rate tolerance
may play a role in the RI determination. Therefore, even if the
BLER is lower than 0.1, it is unlikely to increase the MCS level
by sacrificing BLER to achieve a better system performance.

Fig. 3. Simulation results of the average spectrum efficiency

Fig. 5. MCS distribution of different methods

Fig. 4. Simulation results of the 5th percentile spectrum efficiency

Table 3 MCS index table 2 for PDSCH

The performance of the exhaustive search method is much

lower than the requirements in Fig. 3 and Fig. 4. The proposed
method without the adaptive selection approaches the IMT-2020
evaluation requirements and outperforms the exhaustive search
method. The proposed method with the adaptive selection shows
the highest performance among the three methods and meets the
requirements. Compared to the performance of the exhaustive
search method, the proposed method with the adaptive selection
shows a 74% performance gain in the average spectrum
efficiency and a 130% performance gain in the 5th percentile
spectrum efficiency. Obviously, the adaptive selection
considering the time-varying channel to adjust the RI value
shows a significant improvement. Furthermore, the performance
of the proposed adaptive method is competitive with other
studies [11], as shown in Table 2.

1914
Proceedings, APSIPA Annual Summit and Conference 2021 14-17 December 2021, Tokyo, Japan

Fig. 6 represents the RI distribution. Since the exhaustive

search method and the proposed method without adaptive
selection do not consider the effect of the time-varying channel,
a lower RI value would be selected with a high probability. The
RI distribution (the exhaustive search method, the proposed
method without adaptive selection) is (42.76%, 50.9%) for Rank
1 and (36.93%, 39%) for Rank 2. Obviously, the RI value may
be under-estimated when the error rate tolerance is not
considered. On the other hand, the RI distribution of the
proposed adaptive selection method shows an aggressive RI
determination. The RI distribution is 5.67% for Rank 1, 46.41% Fig. 8. BLER distribution of proposed method without adaptive selection
for Rank 2, 42.83% for Rank 3, and 5.075% for Rank 4. The
spectrum efficiency improvement can be seen in Fig. 3 and Fig. Fig. 9 represents the BLER distribution of the proposed
4. method with the adaptive selection. It shows that the BLER
distribution concentrates more on 0.1, different from the
previous two figures. Obviously, the adaptive RI selection by
considering the BLER performance could achieve a better
system performance.

Fig. 6. RI distribution of different method

B. BLER Distribution
Fig.9. BLER distribution of proposed method with adaptive selection
Fig. 7 represents the BLER distribution in the exhaustive
search method when the UE speed is 3 km/h. Since the channel
variation is not considered, the BLER distribution is mostly
C. Simulation Time
higher than 0.1. This phenomenon denotes that a UE cannot The execution time for the three methods is shown in Table
correctly receive and decodes signals at a desired error rate. The 4. Since the exhaustive search method needs to calculate the
feedback RI may be over-estimates correspondingly. transport block size (TBS) of each RI and to select the RI with
the largest TBS, it requires the longest computational time. The
proposed method by using the MMSE capacity formula to
determine the RI value reduces the execution estimation time
significantly. The proposed method without the adaptive
selection reduces about 56.98% execution time compared to the
exhaustive search method. The proposed method with the
adaptive selection slightly increases the execution time to
determine whether or not to adjust the RI value. Still, it can
obtain a gain of 49.53% compared to the exhaustive search
method. The proposed method with the adaptive selection
reduces the computational time and achieves a better
Fig. 7. BLER distribution of Exhaustive Search Method
performance.
Fig. 8 represents the BLER distribution of the proposed Table 4 Execution time
method without the adaptive selection. It can be observed that Proposed
the error rate is far below 0.1, which means that shifting up the Proposed
Method
RI level could improve the system performance. Exhaustive Method with
欄1 without
Search adaptive
adaptive
selection
selection
Execution 3hr 33m 1hr 31m 1hr 47m
time 24sec 48sec 43sec

1915
Proceedings, APSIPA Annual Summit and Conference 2021 14-17 December 2021, Tokyo, Japan

V. CONCLUSION
D. Mobility Evaluation
By considering the time-varying channel, the BS can use the
Since the proposed adaptive selection method relies on the
BLER to evaluate whether or not to increase the RI value
previous transmission results, the channel variation may affect
according to the feedback RI value from UE. From the
the system performance. Therefore, the performance of the
simulation results of WiSE SLS, it can be found that the
proposed method under different UE speeds, including 10km/h
proposed adaptive method improves the system performance.
and 30km/h, is simulated. Fig. 10 represents the average
The computational time is reduced. Furthermore, the proposed
spectrum efficiency. Fig 11 illustrates the 5th percentile spectrum
adaptive method performs better in a stable scenario, e.g., an
efficiency.
indoor hotspot scenario with low mobility.
Compared the proposed method with the adaptive selection
to the exhaustive search method, the gain of the average
ACKNOWLEDGMENT
spectrum efficiency is 85.98%, and the gain of the 5th percentile
spectrum efficiency is 120.85% when the UE speed is 3km/h. This work was supported in part by the Ministry of Science
The gain of the average spectrum efficiency is 83.67%, and the and Technology (MOST) of Taiwan under Grants MOST109-
gain of the 5th percentile spectrum efficiency is 114.50% when 2218-E-194-007, MOST110-2221-E-194-020, and MOST110-
the UE speed is 10km/h. The gain of the average spectrum 2224-E-305-001 and in part by the Industrial Technology
efficiency is 81.95%, and the gain of the 5th percentile spectrum Research Institute (ITRI), Taiwan.
efficiency is 113.09% when the UE speed is 30km/h.
Compared the proposed method with the adaptive selection REFERENCES
to the proposed method without adaptive selection, the gain of [1] 3GPP TR 38.913, “Study on Scenarios and Requirements for Next
the average spectrum efficiency is 21.84%, and the gain of the Generation Access Technologies.” Jul. 2020.
5th percentile spectrum efficiency is 14.10% when the UE speed [2] 3GPP Technical Report (TR) 38.802 38.214 v14.2.0, “Study on New
Radio Access Technology Physical Layers Aspects (Release 14), Sep.
is 3km/h. The gain of the average spectrum efficiency is 19.33%,
2017.
and the gain of the 5th percentile spectrum efficiency is 12.02% [3] ITU-R Report ITU-R M.2412, “Guidelines for Evaluation of Radio
when the UE speed is 10km/h. The gain of the average spectrum Interface Technologies for IMT-2020,” ITU-R WP 5D, Oct. 2017.
efficiency is 19.96%, and the gain of the 5th percentile spectrum [4] Stefan Schindler, Heinz Mellein, "Assessing a MIMO Channel, ROHDGE
efficiency is 14.61% when the UE speed is 30km/h. Although and SCHWARZ", Feb. 2011.
[5] ShareTechnote,
the performance gain decreases as the mobility speed increases, https://www.sharetechnote.com/html/Handbook_LTE_RI.html
the performance requirements are still satisfied. [6] T. Ghirmai, "Precoder selection and rank adaptation in MIMO-
OFDM," 2012 11th International Conference on Information Science,
Signal Processing and their Applications (ISSPA), Montreal, QC, 2012, pp.
555-560.
[7] COMMRESEARCH, https://www.commresearch.com.tw.
[8] C. K. Jao, C. Y. Wang, T. Y. Yeh, C. C. Tsai, L. C. Lo, J. H. Chen, W. C.
Pao, W. H. Sheen, “WiSE: A System-Level Simulator for 5G Mobile
Networks,” IEEE Wireless Communications, vol. 25, no. 2, pp. 4-7, Apr.
2018.
[9] Harri Holma, Antti Toskala, “LTE Advanced: 3GPP Solution for IMT-
Advanced” Oct. 2012.
[10] David Tse, "Fundamentals of Wireless Communication," 2005.
[11] 3GPP TR 37.910, “Study on self-evaluation towards IMT-2020
submission” Sep. 2019.
[12] 3GPP Technical Specification (TS) 38.214 38.214 v15.8.0, “NR; Physical
Layer Procedure for Data (Release 15),” Dec. 2019.
Fig. 10. Simulation result of average spectrum efficiency in different UE
speed

Fig. 11. Simulation result of 5th percentile spectrum efficiency in different

UE speed

1916