Page 190 - Kaleidoscope Academic Conference Proceedings 2021

2021 ITU Kaleidoscope Academic Conference

In this paper, β norm (t) and φ(t) are chosen as the time 3.1 Threshold comprehensive decision algorithm
correlation characteristics of a wireless channel to
distinguish different speeds. The threshold comprehensive decision algorithm uses the
CIRh(t)to extract a wireless channel characteristic set χ� =
3. ALGORITHM MODEL [δ, P, γ, β, φ], and then make a threshold judgment for each
characteristic, and finally make a comprehensive decision
When the User Equipment (UE) moves slowly, the channel based on the judgment results to distinguish the channel
changes experienced by the signal are relatively slow. At scenarioy� in Table 1. Figure 1 shows the block diagram of
this time, the uplink SRS period can be configured with a the threshold comprehensive decision algorithm.
long period, and the number of uplink DMRS symbols can
also be configured less. When the UE’s speed increases, the δ
channel changes faster, and the base station needs a shorter Threshold decision

δ ≫ η δ
SRS period and more uplink DMRS symbols to track the P Threshold decision
rapid change of the channel. In view of the different delay P ≫ η P
spreads of multipath signals, the base station can use filters computer β Threshold decision Comprehensive
channel
with different window lengths to remove noise and h(t) characteristics β ≫ η β Decision y�
interference, and improve the accuracy of channel γ Threshold decision
estimation. Selecting the UE’s signal under the LOS γ ≫ η γ
channel scenario for positioning will improve the accuracy φ Threshold decision

of positioning. The identification of a LOS/NLOS channel φ ≫ η φ
scenario also provides a novel idea for the optimization of a Figure 1 – Block diagram of wireless channel scenario
MU-MIMO pairing strategy. Considering these applications, recognition using threshold comprehensive decision
there are nine types of wireless channel scenarios to be
identified in this paper, as shown in Table 1. Here, low 3.2 Algorithm model based on neural network
speed means <= 30km/h, medium speed means 30-60km/h,
and high speed means greater than 60km/h. low delay The wireless channel scenario recognition algorithm based
spread means that the multipath delay spread does not on neural networks has two stages: offline training stage
exceed 586ns, and high delay spread means that the and online recognition stage. The training stage is
multipath delay spread does not exceed 2178ns. supervised learning. Firstly, the wireless channel
characteristic set χ = [δ, P, γ, β, φ, y] is extracted using the
Table 1 – Measurement wireless channel scenarios CIR h(t), where y is the channel scenario label. Then the
channel characteristic set χ = [δ, P, γ, β, φ, y] is input into
Label Index Channel Scenario
the neural network for training. After the hyper-parameter
1 LOS, Low Speed tuning, a classifier is obtained as the online recognition
2 LOS, Medium Speed model. The online recognition stage: firstly, get the real-
3 LOS, High Speed time CIR h(t), and then use the same method to obtain the
4 NLOS, Low Delay Spread, Low Speed channel characteristic set, and input it into the trained
5 NLOS, Low Delay Spread Medium Speed neural network model, and lastly perform online
6 NLOS, Low Delay Spread, High Speed recognition to obtain the wireless channel scenario y�.
7 NLOS, High Delay Spread, Low Speed
Train dataset
8 NLOS, High Delay Spread Medium Speed Channel Scenario h(t) Training Phase
9 NLOS, High Delay Spread, High Speed Channel Scenario h(t) computer channel neural
Channel Scenario
characteristics
h(t)
Channel Scenario h(t) χ = [δ, P, γ, β, φ, y] network
...
Considering the hardware cost, a commercial base station Channel Scenario h(t)
generally uses the traditional threshold judgment method Update network parameters
for channel scenario recognition. However, wireless Test dataset Identification Phase
channel recognition is a complicated problem, and the h(t) computer channel y�
h(t)
neural
traditional algorithm may not achieve satisfactory h(t) characteristics network Channel Scenario
performance. Therefore, it is necessary to study AI-based h(t) χ = [δ,P,γ,β,φ]
...
algorithms for channel scenario recognition. In this paper, h(t)
two algorithms are used for wireless channel scenario
recognition: the first is to perform the traditional threshold Figure 2 – Block diagram of wireless channel scenario
comprehensive decision on the extracted channel recognition using neural network
characteristics for channel scenario recognition, and the
second is to use the combination of wireless channel In this paper, the neural network used to distinguish
characteristic sets and neural networks for channel scenario different wireless channel scenarios has 4 layers: one input
recognition. layer, two hidden layers, and one output layer. The neurons
of the hidden layer use the relu activation function. The
neurons of the output layer use the softmax activation

– 128 –