針對毫米波與次太赫茲頻帶之大規模多輸入多輸出通訊的量化混合波束成形器與基於壓縮感知的通道估測方法設計

Design of Quantized Hybrid Beamformer and Compressive Sensing Based Channel Estimation for mmWave/Sub-Thz Band Massive MIMO Communications

張品翔(Ping-Hsiang Chang)

指導教授 ： 闕志達

國立臺灣大學/電機資訊學院/電子工程學研究所/碩士(2022年)

https://doi.org/10.6342/NTU202201362

摘要

隨著現代通訊系統之使用頻帶逐漸往毫米波、甚至太赫茲頻段移動，近年來大規模多輸入多輸出 (massive MIMO) 通訊的應用逐漸受到重視，其在系統中使用大型天線陣列以藉由波束成形 (beamforming) 來對抗高頻通道的強烈通道衰減。其中混合波束得益於其低成本的硬體架構，是實現massive MIMO願景的關鍵技術之一。然而現有的混合波束成形演算法多半具有一些缺點，例如: 未考慮射頻電路精準度之先天硬體限制、假設完美估計的通道等。本論文有兩大主軸，分別是量化混合波束成形係數之設計與 massive MIMO 通道之估計。我們將混合波束成形係數之設計視為一個單層量化神經網路的優化過程，使其可以支援有限精準度的相移器之相位與延遲電路之延遲的量化，並支援多種架構的混和波束成形器。另外我們改善現有壓縮感知 (compressive sensing) 演算法在 massive MIMO 情境中會遇到的瓶頸，並優化其估計精準度與執行複雜度。模擬結果顯示本論文提出之混合波束成形係數設計方法能達到接近全數位波束成形0.5dB內之效能，並能在使用本論文之通道估計方法後，減少因為通道估計誤差所帶來的效能衰減至0.15dB。

本論文的第二章中，我們會說明大規模多輸入多輸出系統的應用，並介紹波束成形與預編碼之數學原理。接著我們會介紹不同波束成形之架構，以及其設計背後的動機與數學模型。

本論文的第三章將介紹我們所使用的通道模型與通道參數之產生方式，我們分別考慮28GHz之毫米波與140GHz之次太赫茲頻段的通道。建構此兩個頻帶的通道模型所需之統計資料則分別來自第五代行動通訊之標準與紐約大學無線中心在次太赫茲頻段的量測結果。另外我們將波束偏斜效應 (beam squint effect) 整合進我們的模型中以捕捉超寬頻系統獨有的傳輸特性。

在第四章中，我們將說明本論文提出之量化相移器與實時延遲之混合波束成形係數設計演算法。另外本論文比較在不同模擬環境、相移器與TTD精準度、硬體架構下，所設計之波束成形係數的頻譜效益。另外我們估計不同架構與量化精準度之波束成形器的面積與功耗效益，以驗證量化後所帶來的優勢，使其可作為實際射頻接收發機設計的參考指標。

在第五章中，我們考慮通道估計誤差對波束成形效能帶來的影響。我們透過早停方法與階層式字典矩陣改善既有基於壓縮感知 (compressed sensing) 的通道估計演算法在massive MIMO情境中複雜度過高與精確度過低的問題。接著我們衡量本論文提出之通道估計方法的正規均方誤差 (normalized mean squared error, NMSE)、複雜度與反饋成本，並與其他文獻和傳統之通道估計方法做比較。最後我們評估本論文第四章設計之混合波束成形演算法，在使用第五章之方法估計之通道作為輸入時，得到的波束成形係數之頻譜效益的衰減情形。

關鍵字

大規模多輸入多輸出 ； 混合波束成形 ； 次太赫茲 ； 實時延遲 ； 波束偏斜 ； 量化感知訓練 ； 壓縮感知

並列摘要

As modern communication system shifts its operating frequency to mmWave or even THz band. The use of massive MIMO (multiple-input-multiple-output) communications has attracted increasing attention. Massive MIMO systems exploit massive antenna arrays to combat the severe path loss in the high-frequency band. Hybrid beamforming has since become one of the key technologies to achieve the vision of massive MIMO for its low-cost hardware architecture. However, most existing hybrid beamforming algorithms in the literature have some drawbacks. For example, they did not consider the limited resolution of RF hardware and perfect CSI (channel state information) is often assumed in these references. Toward this end, there are mainly two topics in this thesis, one is the design of quantized hybrid beamforming weights, and the other is the estimation of massive MIMO channels. We model the design of hybrid beamforming coefficients as a single-layer quantized neural network optimization problem, allowing it to quantify RF phase shifters’ phase and the amount of delay for the TTD (true time delay) circuits. The proposed algorithm can support different hybrid beamforming architectures with different resolutions. On the other hand, we improve the bottleneck faced by the existing compressed-sensing-based channel estimation algorithms in the massive MIMO scenario and optimize its complexity and estimation accuracy. Simulation shows that the proposed hybrid beamforming algorithm can achieve spectral efficiency close to that of all-digital beamforming with less than 0.5dB performance loss while reducing the performance degradation brought by estimation error to less than 0.15dB when proposed estimation algorithm is applied.

In chapter 2, we will first introduce the application of massive MIMO systems. Then we review the basic concepts and the mathematical formulations of beamforming and precoding. Finally, we will introduce the motivation behind different beamforming architectures and their corresponding mathematical models.

In chapter 3, we explain the channel model used throughout this thesis and the corresponding procedures to generate each channel parameter. We consider both the mmWave channel at 28GHz and the sub-THz channel at 140GHz. The statistical data for channel generation comes respectively from the 5G standard and the reported sub-THz measurement result from NYU WIRELESS. In addition, we integrate the beam squint effect in our channel model to capture the unique characteristics of wideband communications.

Chapter 4 introduces the proposed quantized-phase and quantized-delay hybrid beamforming algorithm and compares the spectral efficiency of designed beamforming weights in different simulation environments, phase shift and TTD circuit resolutions, and hardware architectures. Besides, we estimate the area and power efficiency of different hybrid beamforming architectures with different resolutions to validate quantization’s advantage and provide a reference for RF transceiver design.

Chapter 5 considers the impact of estimation error on the beamforming performance. We improve existing compressed-sensing-based channel estimation algorithms’ high complexity and low accuracy when applied in the massive MIMO scenarios, through a hierarchical dictionary and early termination method. Next, we evaluate the NMSE (normalized mean squared error), algorithmic complexity and the feedback overhead of the proposed algorithm. Finally, we estimate the spectral efficiency degradation of hybrid beamforming coefficients derived from the algorithm proposed in chapter 4 when using the channel estimated by the method proposed in chapter 5 as the input.

並列關鍵字

massive MIMO ； hybrid beamforming ； sub-THz ； true time delay ； beam squint ； quatization-aware training ； compressed sensing

參考文獻

[1] Ericsson, "Mobile data traffic outlook." ericsson.com. https://www.ericsson.com/en/reports-and-papers/mobility-report/dataforecasts/mobile-traffic-forecast (accessed Apr. 16, 2022).

[2] M. Latva-Aho and K. Leppänen, “Key drivers and research challenges for 6G ubiquitous wireless intelligence,” 6G Flagship, Univ. Oulu, Finland, Tech. Rep., 2019.

[3] C. Han, J. M. Jornet, and I. Akyildiz, “Ultra-massive MIMO channel modeling for

graphene-enabled terahertz-band communications,” in Proc. IEEE VTC Spring, Jun. 2018, pp. 1-5.

[4] J. M. Jornet and I. F. Akyildiz, "Graphene-based Plasmonic Nano-Antenna for Terahertz Band Communication in Nanonetworks," IEEE J. Sel. Areas Commun., vol. 31, no. 12, pp. 685-694, December 2013.