Multi-User Delay Alignment Modulation for Millimeter Wave Massive MIMO

Recently, delay alignment modulation (DAM) was proposed as a novel technique to tackle the inter-symbol interference (ISI) issue, without relying on conventional techniques like channel equalization or multi-carrier transmission [1]. Specifically, by leveraging the super spatial resolution of large antenna arrays [2] and the inherent multi-path sparsity of high frequency channels like millimeter wave (mmWave)[3] and Terahertz channels, DAM enables manipulable channel delay spread by means of delay compensation and path-based beamforming [1]. As a result, all the multi-path signal components may reach the receiver concurrently and constructively, rather than causing the detrimental ISI. This renders DAM resilient to the time-dispersive channel for more efficient single- or multi-carrier transmissions. Some preliminary works on DAM are presented in [4, 5, 6, 7, 8]. For example, in [4], an efficient channel estimation method was developed for DAM communication. By combining DAM with orthogonal frequency division multiplexing (OFDM), a novel DAM-OFDM scheme was proposed in [5], which may outperform OFDM in terms of spectral efficiency, bit error rate (BER), and peak-to-average-power ratio (PAPR). The investigations of DAM for integrated sensing and communication (ISAC) and multiple-intelligent reflecting surfaces (IRSs) aided communication were investigated in [6], [7] and [8], respectively.

DAM is significantly different from the existing ISI-mitigation techniques developed over the fast few decades, such as channel equalization and multi-carrier OFDM transmission. For example, typical time-domain equalization techniques include time reversal (TR)[9, 10, 11] and channel shortening[12, 13]. Specifically, TR mainly treats the multi-path channel as an intrinsic matched filter[9, 10, 11] and addresses the ISI issue via the rate back-off technique[9, 10]. In [12] and [13], channel shortening technique was proposed by applying a short time-domain finite impulse response (FIR) filter to shorten the effective channel impulse response in order to avoid the use of long cyclic prefix (CP). Besides, OFDM is the dominating wideband communication technology, though it suffers from well-known practical issues like carrier frequency offset (CFO), high PAPR, and the severe out-of-band (OOB) emission[14]. By contrast, DAM is a spatial-delay processing technique, which utilizes the high spatial resolution and multi-path sparsity of mmWave massive multiple-input multiple-output (MIMO) systems, while circumventing the aforementioned issues suffered by conventional channel equalization or multi-carrier communication. It is worth noting that relevant techniques for delay spread reduction were studied in [15] and [16]. However, neither of them exploits the high spatial resolution or multi-path sparsity of mmWave massive MIMO systems to completely eliminate ISI. In addition, both our prior work [17] and [18] apply delay compensation in the mmWave MIMO systems, but they were designed only for the special lens MIMO or hybrid beamforming architectures.

Note that the existing works[1, 4, 5, 6, 7, 8] on DAM only focus on the single user scenario. In this paper, we extend the DAM technique to the multi-user mmWave massive MIMO systems. To gain the essential insights, we first provide an asymptotic analysis by assuming that the number of BS antennas $M_{t}$ is much larger than the total number of channel multi-paths $L_{\mathrm{tot}}$. In this case, we show that the time-dispersive multi-user channel can be transformed into ISI-free and inter-user interference (IUI)-free additive white Gaussian noise (AWGN) channels with the simple delay pre-compensation and per-path-based MRT beamforming. Then for the general scenario with given finite number of BS antennas, the multi-user DAM design is investigated by considering three classical transmit beamforming strategies in a per-path basis, i.e., MRT, zero-forcing (ZF) and regularized zero-forcing (RZF). To be specific, it is shown than when $M_{t}\geq L_{\mathrm{tot}}$, both the ISI and IUI can be completely eliminated with the path-based ZF beamforming. Furthermore, the low complexity path-based MRT beamforming and the more general path-based RZF beamforming are respectively developed for multi-user DAM with tolerable residual ISI and IUI. Finally, simulation results are provided to demonstrate the superiority of DAM to the benchmarking single-carrier ISI mitigation technique that only uses the strongest path of each user.

As shown in Fig. 1, we consider a multi-user mmWave massive MIMO downlink communication system, where the BS equipped with $M_{t}\gg 1$ antennas serves $K$ single-antenna user equipments (UEs). In multi-path environment, the baseband discrete-time channel impulse response of UE $k$ for each channel coherence block can be expressed as

where $\boldsymbol{\bf{h}}_{kl}\in\mathbb{C}^{M_{t}\times 1}$ denotes the channel coefficient vector for the $l$th multi-path of UE $k$, $L_{k}$ is the number of temporal-resolvable multi-paths of UE $k$, and $n_{kl}$ denotes its discretized delay. Let $n_{k,\text{max}}=\max\limits_{1\leq l\leq L_{k}}n_{kl}$ and $n_{k,\text{min}}=\min\limits_{1\leq l\leq L_{k}}n_{kl}$ denote the maximum and minimum delays of UE $k$, respectively.

Let $s_{k}[n]$ be the independent and identically distributed (i.i.d.) information-bearing symbols of UE $k$ with normalized power $\mathbb{E}[|s_{k}[n]|^{2}]=1$. By extending the DAM technique proposed in [1] to the multi-user scenario, the transmitted signal by the BS for multi-user DAM is

where $\boldsymbol{\bf{f}}_{kl^{\prime}}\in\mathbb{C}^{M_{t}\times 1}$ denotes the per-path based transmit beamforming vector associated with multi-path $l^{\prime}$ of UE $k$, and $\kappa_{kl^{\prime}}$ is the deliberately introduced delay pre-compensation, which is set as $\kappa_{kl^{\prime}}=n_{k,\text{max}}-n_{kl^{\prime}}\geq 0,\forall l^{\prime}=1,...,L_{k}$. The block diagram for multi-user DAM is given in Fig. 2. By using the fact that $s_{k}[n]$ is independent across different $n$ and $k$ as well as $\kappa_{kl^{\prime}}\neq\kappa_{kl},\forall l\neq l^{\prime}$, it can be shown that the transmit power of the BS is

where $P$ is the maximum allowable transmit power.

Based on (1) and (2), the received signal of UE $k$ is

where $*$ denotes linear convolution, and $z_{k}[n]\sim\mathcal{CN}(0,\sigma^{2})$ is the AWGN. It is observed that if UE $k$ is locked to the delay $n_{k,\text{max}}$, then the first term is the desired signal, and the second and third terms are the ISI and IUI, respectively.

Fortunately, we show that in mmWave massive MIMO systems when the number of BS antennas $M_{t}$ is much larger than the total number of multi-paths $L_{\mathrm{tot}}=\sum_{k=1}^{K}L_{k}$, both the ISI and IUI in (4) asymptotically vanish with the simple path-based MRT beamforming. To this end, we first study the correlation property of the channel vectors $\boldsymbol{\bf{h}}_{kl},k=1,...,K,l=1,...,L_{k}$. For ease of exposition, the basic uniform linear array (ULA) with adjacent elements separated by half wavelength is considered, for which $\boldsymbol{\bf{h}}_{kl}$ in (1) can be written as

where $\alpha_{kl}$ denotes the complex-valued path gain for the $l$th multi-path of UE $k$, and $\theta_{kl}$ and $\boldsymbol{\bf{a}}(\theta_{kl})\in\mathbb{C}^{M_{t}\times 1}$ denote the angle of departure (AoD) and the transmit array response vector of multi-path $l$ of UE $k$, respectively. Note that for ease of the asymptotical analysis, we assume that each temporal-resolvable multi-path corresponds to one AoD. It follows from [5] that as long as the multi-paths correspond to distinct AoDs, i.e., $\theta_{kl}\neq\theta_{k^{\prime}l^{\prime}},\forall k\neq k^{\prime}$ or $l\neq l^{\prime}$, the transmit array response vectors are asymptotically orthogonal when $M_{t}\gg L_{\mathrm{tot}}$, i.e.,

Consider the low-complexity path-based MRT beamforming $\boldsymbol{\bf{f}}_{kl^{\prime}}=\xi_{k}\sqrt{p_{k}}\boldsymbol{\bf{h}}_{kl^{\prime}}$, where $\xi_{k}=1/\sqrt{\sum_{j=1}^{L_{k}}\|\boldsymbol{\bf{h}}_{kj}\|^{2}}$ is the normalization factor and $p_{k}$ denotes the power allocated to UE $k$. In this case, by scaling the signal in (4) with $\xi_{k}$, we have

Thanks to the asymptotically orthogonal property in (6), when $M_{t}\gg L_{\mathrm{tot}}$, we have

and

With (8) and (9), and by dividing both sides of (II) with $\xi_{k}$, the resulting signal for UE $k$ reduces to

It is observed from (10) that the resulting signal of UE $k$ only includes the symbol sequence $s_{k}[n]$ with one single delay $n_{k,\text{max}}$, while achieving the multiplicative gain contributed by all the $L_{k}$ multi-paths. As a result, the original time-dispersive multi-user interfering channel has been transformed to $K$ parallel ISI- and IUI-free AWGN channels, without relying on the conventional techniques like channel equalization or multi-carrier transmission. Moreover, the signal-to-noise ratio (SNR) in (10) is given by $\gamma_{k}^{\text{MRT}}={p_{k}\|\bar{\boldsymbol{\bf{h}}}_{k}\|^{2}}/{\sigma^{2}}$, where $\bar{\boldsymbol{\bf{h}}}_{k}=[\boldsymbol{\bf{h}}_{k1}^{H},...,\boldsymbol{\bf{h}}_{kL_{k}}^{H}]^{H}\in\mathbb{C}^{M_{t}L_{k}\times 1}$. The optimal power allocation to maximize the asymptotic sum rate can be obtained by the classical water-filling (WF) power allocation.

In this section, we consider the practical scenario with given finite number of antennas $M_{t}$, where three classical precoding schemes, namely MRT, ZF and RZF are studied on the per-path basis to cater for multi-user DAM transmission.

To derive the signal-to-interference-plus-noise ratio (SINR) for the general signal in (4), the symbols need to be properly grouped by considering the delay differences of the signal components[1]. To this end, let $\Delta_{kl,k^{\prime}l^{\prime}}=n_{kl}-n_{k^{\prime}l^{\prime}}$ denote the delay difference between multi-path $l$ of UE $k$ and multi-path $l^{\prime}$ of UE $k^{\prime}$. Then for a given UE pair $(k,k^{\prime})$, we have $\Delta_{kl,k^{\prime}l^{\prime}}\in\{\Delta_{kk^{\prime},\min},\Delta_{kk^{\prime},\min}+1,...,\Delta_{kk^{\prime},\max}\}$, where $\Delta_{kk^{\prime},\max}=n_{k,\text{max}}-n_{k^{\prime},\text{min}}$ and $\Delta_{kk^{\prime},\min}=n_{k,\text{min}}-n_{k^{\prime},\text{max}}$. Then for each UE pair $(k,k^{\prime})$ and delay difference $i$ with $i\in\{\Delta_{kk^{\prime},\min},\Delta_{kk^{\prime},\min}+1,...,\Delta_{kk^{\prime},\max}\}$, we have the following effective channel

Therefore, (4) can be equivalently written as

The resulting SINR is given in (13), shown at the top of the next page, where we have defined $\bar{\boldsymbol{\bf{f}}}_{k}=[\boldsymbol{\bf{f}}_{k1}^{H},...,\boldsymbol{\bf{f}}_{kL_{k}}^{H}]^{H}\in\mathbb{C}^{M_{t}L_{k}\times 1},\bar{\boldsymbol{\bf{g}}}_{kk^{\prime}}[i]=[\boldsymbol{\bf{g}}_{kk^{\prime}1}^{H}[i],...,\boldsymbol{\bf{g}}_{kk^{\prime}L_{k^{\prime}}}^{H}[i]]^{H}\in\mathbb{C}^{M_{t}L_{k^{\prime}}\times 1}$, $\boldsymbol{\bf{G}}_{kk^{\prime}}=[\bar{\boldsymbol{\bf{g}}}_{kk^{\prime}}[\Delta_{kk^{\prime},\min}],...,\bar{\boldsymbol{\bf{g}}}_{kk^{\prime}}[\Delta_{kk^{\prime},\max}]]\in\mathbb{C}^{M_{t}L_{k^{\prime}}\times{\Delta_{kk^{\prime},\text{span}}}}$, with $\Delta_{kk^{\prime},\text{span}}=\Delta_{kk^{\prime},\max}-\Delta_{kk^{\prime},\min}$.