Page 188 - Kaleidoscope Academic Conference Proceedings 2021

2021 ITU Kaleidoscope Academic Conference

channel estimation according to the delay spread of the 2.1 Frequency domain fading factor
wireless channel multipath, and then improve the receiving
performance. If the base station can distinguish the different The time dispersion and frequency domain selective fading
speeds, the period of the uplink Sounding Reference Signal of the wireless channel are all produced by the
(SRS) can be adapted according to the user’s speed. The superposition of multipath signals with different time
Demodulation Reference Signal (DMRS) also can adjust delays. The two effects exist at the same time, the time
adaptively the number of pilots according to the speed. The dispersion is reflected in the time domain, and the
recognition of LOS/NLOS channel scenarios can be used frequency selective fading is reflected in the frequency
for positioning or as a priority reference for Multi-User domain. The frequency selective fading is to filter the
Multiple-Input Multiple-Output (MU-MIMO) user pairing. transmitted signal, and the different frequency components
Then this paper mainly studies the wireless channel of the signal have different fading amplitudes. When the
scenario recognition in 4G or 5G dense urban areas and transmitted signal bandwidth is narrow enough, all
rural or suburban areas. The neural network is used to frequency components almost experience the same fading,
realize wireless channel scenario recognition. The main and which is non-frequency selective fading. When the
contributions are summarized as follows: signal bandwidth is relatively large, the different frequency
components have different attenuation coefficients, which
1. Extract key wireless channel characteristics, including is frequency selective fading.
frequency domain fading characteristic, multipath power
delay distribution, channel peak response ratio in time In this paper, the power variance of the channel impulse
domain and time correlation characteristics. These response in the frequency domain is used to characterize the
characteristics are easy to calculate, which has high frequency selective fading. Assuming that in the 5G NR
engineering application value. system, the transmitted signal occupies N RB resource blocks
in the frequency domain. After the channel estimation in
2. Propose a wireless channel scenario recognition the receiver, the CIR on subcarrier k at the moment t is
framework based on neural networks, which can not only h(k, t). There are 6 pilot subcarriers on one Resource Block
distinguish LOS/NLOS scenarios, but also identify different (RB), and then the signal power on resource block RB at
speeds and different multipath delay spreads. the moment t is calculated by Equation (1). i

3. The performance of the traditional threshold algorithm P(RB , t) = ∑ 5 |h(k, t)| （1）
1
2
and the neural network classifier is compared, and the i 6 k=0
recognition results of the neural network classifier under
different channel feature composition structures and At the momentt, the average power of the signal on the
different Signal-to-Noise Ratio (SNR) are given. occupied frequency domain bandwidth is calculated by
Equation (2).
2. WIRELESS CHANNEL FEATURE
EXTRACTION P �(t) = 1 ∑ N RB −1 P(RB , t) （2）
i
N RB RB i =0
As mentioned in Section 1, the receiving algorithm can At the moment t , the normalized power variance is
make adaptive adjustments based on the identified wireless calculated by Equation (3).
channels to improve system performance. The wireless
channel scenarios that are mainly studied and distinguished 1 N RB −1 2
�
in this paper are: (1) LOS or NLOS, (2) Different speeds: δ(t) = N RB ∑ RB i =0 �P(RB , t) − P �(t)� P �(t) （3）
i
low speed, medium speed and high speed, (3) Multipath
delay spread: low and high. 2.2 Multipath power delay distribution
Choosing information-rich, discriminative features is the Time dispersion and frequency domain selective fading
key to an effective classification algorithm. In this section, exist simultaneously. The frequency domain feature is
Channel Impulse Response (CIR) is used to calculate extracted in Section 2.1, and the time dispersion feature is
wireless channel characteristics. Because only LOS/NLOS, extracted in this section. Time dispersion is usually
different multipath delay spread and different speeds are characterized by the average delay spread and the root
distinguished in this paper, there is no distinction in spatial mean square delay, as in [11] and [13]. However, in order
dimension, so we choose frequency domain fading factor δ, to obtain the average delay spread and the root mean square
multipath power delay distribution P, channel power peak delay, it is necessary to estimate accurately the delay and
response ratio γ in the time domain, channel time the power of each propagation path, and which is a
correlation value β and its angle φ as the wireless channel challenge for a commercial communication system. In this
characteristics; the characteristic set of the wireless channel paper, a multipath power delay distribution is calculated
can be defined as χ = [δ, P, γ, β, φ, y] , where y is the based on the CIR in the time domain. There is no need to
classification label of the wireless channel scenario. separate each propagation path, which reduces the
complexity of implementation.

– 126 –