---
title: 'WideNLNet-CA: Adaptive CSI Feedback'
url: https://www.emergentmind.com/topics/widenlnet-ca
type: topic
---

# WideNLNet-CA: Adaptive CSI Feedback

WideNLNet-CA is a rate-adaptive deep learning framework for channel state information (CSI) feedback in wideband near-field extremely large-scale MIMO (XL-MIMO) systems. It is introduced for settings in which near-field spherical wave propagation and frequency-dependent beam split jointly complicate CSI representation and compression. The framework combines a lightweight encoder–decoder with multi-stage downsampling and upsampling, computationally efficient residual blocks, and a compression ratio adaptive module (CAM) that conditions feature selection on the target feedback rate. Its stated objective is to support variable compression ratios with a single model while preserving reconstruction quality, fast inference, and low model storage requirements in wideband near-field XL-MIMO feedback [2508.00626].

## 1. Signal model and wideband near-field feedback problem

The underlying system is a wideband XL-MIMO downlink with \(N\) base-station antennas arranged as a ULA and \(M\) OFDM subcarriers. The spatial–frequency–domain CSI is represented as
\[
\tilde{\mathbf H} = [\mathbf h_1,\dots,\mathbf h_M]^H \in \mathbb{C}^{M\times N}.
\]
A 2D DFT is then applied to obtain the angular-delay-domain tensor
\[
\mathbf H = \mathbf F_d^H\,\tilde{\mathbf H}\,\mathbf F_a \in \mathbb{C}^{M\times N},
\]
which is reshaped into a real-valued input
\[
\mathbf H_{\mathrm{ts}}\in\mathbb R^{N\times M\times 2}
\]
by splitting the real and imaginary parts.

Two propagation effects define the problem setting. First, under spherical wavefronts, the phase at antenna \(n\) on subcarrier \(m\) is
\[
\phi_{l,m}^{(n)}=-\tfrac{2\pi}{\lambda_m}(r_{l}^{(n)}-r_l),
\]
which is explicitly not linear in \(n\). Second, beam split causes a frequency-dependent array response and therefore spatial dispersion across subcarriers. In combination, these effects make the CSI structure more difficult to compress with fixed-rate or narrowband-oriented feedback schemes.

The feedback compression ratio is defined as
\[
r = \beta = \frac{K_{\rm code}}{2NM},
\]
where \(K_{\rm code}\) is the codeword dimension. A central design target of WideNLNet-CA is support for variable \(\beta\) with one model. This directly situates the framework in the class of rate-adaptive CSI feedback methods rather than single-compression-ratio autoencoders.

## 2. Encoder–decoder formulation

The encoder \(f_{\rm en}(\cdot;\Phi)\) maps \(\mathbf H_{\mathrm{ts}}\) to a latent code through a hierarchical convolutional pipeline. It begins with an initial \(3\times 3\) convolution,
\[
F_0 = \mathrm{Conv}_{3\times3}\bigl(\mathbf H_{\mathrm{ts}};W_0,b_0\bigr)\in\mathbb R^{N\times M\times C_1},
\]
followed by three downsampling blocks indexed by \(i=1,2,3\). Each block applies a strided \(3\times 3\) convolution with stride \(2\),
\[
\tilde F_i = \mathrm{Conv}_{3\times3,s=2}(F_{i-1};W_i,b_i),
\]
and then a LightResBlock,
\[
F_i=\mathrm{LRB}(\tilde F_i).
\]
The resulting feature map \(F_3\) has size \(\tfrac{N}{8}\times\tfrac{M}{8}\times C_4\) and serves as the input to the CAM.

The decoder \(f_{\rm de}(\cdot;\Psi)\) mirrors the encoder. A \(1\times 1\) convolution first restores channels,
\[
G_0=\mathrm{Conv}_{1\times1}( \hat z_{\rm CAM\_out};W'_0,b'_0),
\]
after which three upsampling blocks apply transposed convolutions with stride \(2\) and LightResBlocks. Reconstruction ends with a final \(3\times 3\) convolution and Sigmoid,
\[
\hat{\mathbf H}_{\mathrm{ts}} = \sigma\bigl(\mathrm{Conv}_{3\times3}(G_3;W'_f,b'_f)\bigr)\in[0,1]^{N\times M\times2}.
\]

This encoder–decoder organization is explicitly lightweight and multi-stage. The stated rationale is that multi-stage downsampling and upsampling capture complex multi-scale channel features while keeping overhead low. A plausible implication is that the architecture is intended to preserve enough spatial–frequency structure for reconstruction without resorting to a large monolithic latent bottleneck.

## 3. LightResBlock and computational efficiency

The residual unit used throughout the network is the LightResBlock (LRB), which employs depthwise-separable convolutions. For input \(X\), it is defined as
\[
\begin{aligned}
U &= \sigma\bigl(\mathrm{DWConv}_{3\times3}(X;W_1)\bigr),\\
V &= \sigma\bigl(\mathrm{PWConv}_{1\times1}(U;W_2)+b_2\bigr),
\end{aligned}
\]
with skip connection
\[
X_{\rm out} = X + V.
\]

The design objective of the LRB is stated directly: it captures multi-scale local features while keeping parameter count low. In WideNLNet-CA, this block is used after each strided convolution in the encoder and after each transposed convolution in the decoder. The resulting pattern makes the residual path a structural component of both compression and reconstruction.

Within the framework, efficiency is therefore not only a property of model size but also of operator choice. The use of depthwise-separable convolutions places the LRB in a specifically lightweight regime, which aligns with the reported emphasis on fast inference and low model storage requirements. This suggests that WideNLNet-CA is intended for scenarios where feedback quality and deployment constraints must be balanced simultaneously.

## 4. Compression Ratio Adaptive Module

The compression ratio adaptive module is the mechanism that makes WideNLNet-CA rate-adaptive. Starting from encoder output \(F_3\in\mathbb R^{\frac N8\times\frac M8\times C_b}\), the model applies a \(3\times 3\) convolution to obtain \(Z_{\rm conv}\). At each spatial location, a feature vector \(z\in\mathbb R^{C_b}\) is processed by a feature-importance estimator
\[
s = f_{\rm FI}(z\,; \theta)\in\mathbb R^{C_b},
\]
implemented via four fully connected layers with intermediate adaptation conditioned on the target channel count
\[
C_t=128\,\beta.
\]

The adaptation mechanism is itself explicit. At adaptation module \(j\), the modulation is
\[
m_j = \mathrm{Sigmoid}\bigl(W^{(3)}_j (\mathrm{ReLU}(W^{(2)}_j(\mathrm{ReLU}(W^{(1)}_j C_t + b^{(1)}_j)) + b^{(2)}_j)) + b^{(3)}_j\bigr).
\]
These cascaded modulations operate on the FC stages and yield the final per-channel importance map
\[
\alpha = \mathrm{Sigmoid}(s)\in(0,1)^{\frac N8\times\frac M8\times C_b}.
\]

Feature selection proceeds in two stages. First, \(\alpha\) is aggregated spatially to obtain global channel scores \(\bar\alpha\in\mathbb R^{C_b}\). Then the top \(C_t\) channels are selected to form a binary mask \(M\in\{0,1\}^{C_b}\). The feature modulation and masking are
\[
Z_{\rm mod} = Z_{\rm conv}\odot\alpha,\quad
Z_{\rm CAM\_out} = Z_{\rm mod}\odot M,
\]
which yields exactly \(C_t\) active feature maps as the codeword \(\mathbf s\).

The significance of CAM is that a single WideNLNet-CA instance supports multiple \(\beta\) through conditioning on \(C_t\). The framework does not claim to remove the rate–distortion trade-off. On the contrary, its stated trade-off is that lower CR, corresponding to smaller \(\beta\), leads to fewer channels selected and higher NMSE; the role of CAM is to ensure graceful degradation rather than invariance across compression ratios.

## 5. Training protocol and implementation configuration

Training minimizes the mean squared error over a training set of size \(D\):
\[
\mathcal L_{\rm MSE}
= \frac1D\sum_{i=1}^D\bigl\|\hat{\mathbf H}_{\mathrm{ts}}^{(i)} - \mathbf H_{\mathrm{ts}}^{(i)}\bigr\|_F^2.
\]
Multi-rate training is performed by selecting a random \(\beta\in\{1/4,1/16,1/64\}\) for each batch and setting \(C_t\) accordingly. No additional regularization terms were required beyond batch normalization and weight decay inherent in Adam.

The implementation details are fully specified. The dataset consists of 45,000 spherical-model channels, divided into 35,000 training samples, 5,000 validation samples, and 5,000 test samples. The configuration uses \(N=M=256\), a \(100\,\mathrm{GHz}\) carrier, and \(10\,\mathrm{GHz}\) bandwidth. Optimization uses Adam with learning rate \(3\times10^{-4}\), batch size \(64\), and \(200\) epochs. Feature depths are \(C_1=C_2=64\), \(C_3=C_4=128\), CAM depth \(C_b=32\), and latent \(K=64\). The implementation framework is PyTorch on an NVIDIA RTX 4090 GPU and an Intel Xeon w7-3465X CPU.

These details indicate that WideNLNet-CA is trained as a single shared model across multiple feedback rates rather than as a collection of separately optimized single-rate models. A plausible implication is that the CAM is not an auxiliary add-on but the primary mechanism through which rate adaptivity is learned during optimization.

## 6. Empirical behavior, comparative results, and significance

The reported evaluation emphasizes reconstruction accuracy, bandwidth robustness, and deployment efficiency [2508.00626]. At compression ratio \(1/64\), WideNLNet-CA achieves \(\mathrm{NMSE}\approx -11.8\,\mathrm{dB}\), compared with ExtendNLNet at approximately \(-5\,\mathrm{dB}\) and BSPD at greater than \(-2\,\mathrm{dB}\). Cosine similarity remains greater than \(0.95\) even at high compression. Under bandwidth variation at \(\mathrm{CR}=1/16\), as \(B\) increases from \(1\,\mathrm{GHz}\) to \(10\,\mathrm{GHz}\), WideNLNet-CA degrades by less than \(1\,\mathrm{dB}\) NMSE, whereas baselines lose more than \(3\,\mathrm{dB}\).

The reported complexity and latency results are as follows:

| Method | Parameters | Inference time |
|---|---:|---:|
| BSPD | 1.07 M | 130 ms |
| ExtendNLNet / ENL | 8.39 M | 2.51 ms |
| WideNLNet / WNL | 1.47 M | 1.57 ms |
| WideNLNet-CA / WNL-CA | 1.52 M | 1.64 ms |

Only WideNLNet-CA is stated to use one model for all compression ratios. This is significant because the rate-adaptive claim is tied not merely to reconstruction quality but also to model consolidation. The framework’s contribution is therefore twofold: it introduces a rate-conditioned feature-importance mechanism and demonstrates that this mechanism can operate with minimal additional overhead relative to WideNLNet, while maintaining what the source describes as flexible, accurate, and efficient CSI feedback with state-of-the-art performance.

A common misunderstanding would be to interpret rate adaptivity as implying compression-invariant accuracy. The reported trade-off rejects that interpretation: lower compression ratios correspond to fewer selected channels and higher NMSE. The empirical claim is instead that the degradation is graceful, and that this occurs within a single trained model across a broad range of feedback rates.

Source: https://www.emergentmind.com/topics/widenlnet-ca