Ferkans — Interactive Telecom Tutor

Why Uplink Detection Matters

In a massive MIMO system, all the heavy signal processing happens at the base station (BS) — and in the uplink, the BS must detect the symbols transmitted by $K$ users from a single received vector. This is fundamentally a multiple-access problem: the users share the same time-frequency resource, and the BS exploits its $N_t$ antennas to separate them spatially.

The choice of detection algorithm determines the tradeoff between computational complexity and spectral efficiency. Simple linear receivers (MRC, ZF) become near-optimal in the massive MIMO regime $N_t \gg K$ , which is precisely the operational simplification that makes massive MIMO attractive. But when the antenna-to-user ratio is moderate, or when the channel is ill-conditioned, more sophisticated receivers (MMSE, MMSE-SIC) are required to approach the information-theoretic limits.

Definition:
Massive MIMO Uplink System Model

Consider a single-cell uplink with $K$ single-antenna users transmitting to a BS equipped with $N_t$ antennas. The received signal at the BS is

$\mathbf{y} = \sum_{k=1}^{K} \mathbf{h}_k x_k + \mathbf{w} = \mathbf{H} \mathbf{x} + \mathbf{w},$

where:

$\mathbf{h}_k \in \mathbb{C}^{N_t}$ is the channel vector from user $k$ ,
$x_k$ is the transmitted symbol of user $k$ with power $P_k = \mathbb{E}[|x_k|^2]$ ,
$\mathbf{x} = [x_1, \ldots, x_{K}]^T \in \mathbb{C}^{K}$ is the stacked transmit vector,
$\mathbf{H} = [\mathbf{h}_1, \ldots, \mathbf{h}_{K}] \in \mathbb{C}^{N_t \times K}$ is the channel matrix,
$\mathbf{w} \sim \mathcal{CN}(\mathbf{0}, \sigma^2 \mathbf{I}_{N_t})$ is AWGN noise.

We assume $N_t \geq K$ (more BS antennas than users), which is the standard massive MIMO operating regime.

In TDD systems, the BS obtains $\mathbf{H}$ (or an estimate $\hat{\mathbf{H}}$ ) from uplink pilot training using channel reciprocity. The detection algorithms in this chapter assume knowledge of $\mathbf{H}$ ; the impact of imperfect CSI is treated via the UatF bound (MIMO Ch. 4).

Definition:
Linear Detector

A linear detector applies a combining matrix $\mathbf{G} \in \mathbb{C}^{N_t \times K}$ to the received signal to produce soft estimates of all $K$ users:

$\hat{\mathbf{x}} = \mathbf{G}^H \mathbf{y}.$

The $k$ -th element $\hat{x}_k = \mathbf{g}_k^H \mathbf{y}$ is the soft estimate of user $k$ 's symbol, where $\mathbf{g}_k$ is the $k$ -th column of $\mathbf{G}$ . The detected symbol (hard decision) is then obtained by slicing $\hat{x}_k$ to the nearest constellation point.

Linear detection has complexity $\mathcal{O}(N_t K)$ per symbol vector (matrix-vector multiply) plus a one-time $\mathcal{O}(K^{2} N_t)$ cost for computing $\mathbf{G}$ .

Definition:
Post-Detection SINR

For a linear detector with combining vector $\mathbf{g}_k$ , the post-detection SINR of user $k$ is

$\text{SINR}_k = \frac{P_k |\mathbf{g}_k^H \mathbf{h}_k|^2} {\sum_{j \neq k} P_j |\mathbf{g}_k^H \mathbf{h}_j|^2 + \sigma^2 \|\mathbf{g}_k\|^2}.$

The achievable rate for user $k$ is $R_k = \log_2(1 + \text{SINR}_k)$ bits/s/Hz.

Definition:
Gram Matrix

The Gram matrix of the channel is

$\mathbf{G}_{\text{ram}} \triangleq \mathbf{H}^{H} \mathbf{H} \in \mathbb{C}^{K \times K},$

with entries $[\mathbf{G}_{\text{ram}}]_{k,j} = \mathbf{h}_k^H \mathbf{h}_j$ . The diagonal elements $\mathbf{h}_k^H \mathbf{h}_k = \|\mathbf{h}_k\|^2$ are the channel gains, and the off-diagonal elements measure inter-user spatial correlation. The Gram matrix is central to ZF and MMSE detection: both require inverting (or regularized inverting) $\mathbf{H}^{H} \mathbf{H}$ .

In the massive MIMO regime, $\frac{1}{N_t} \mathbf{H}^{H} \mathbf{H} \to \text{diag}(\beta_1, \ldots, \beta_{K})$ by favorable propagation, where $\beta_k$ is the large-scale fading coefficient of user $k$ . This approximate diagonalization is why linear receivers become near-optimal.

Linear Detector

A receiver that forms the symbol estimate as a linear function $\hat{\mathbf{x}} = \mathbf{G}^H \mathbf{y}$ of the received signal. Common choices: MRC, ZF, MMSE.

SINR (Signal-to-Interference-plus-Noise Ratio)

The ratio of desired signal power to the sum of interference power from other users and thermal noise power, measured at the detector output. Determines achievable rate via $R = \log_2(1 + \text{SINR})$ .

Gram Matrix

The matrix $\mathbf{H}^{H} \mathbf{H}$ whose $(k,j)$ -th entry is the inner product $\mathbf{h}_k^H \mathbf{h}_j$ of the channel vectors. Its condition number governs the noise enhancement in ZF detection.

Why This Matters: Uplink Detection in 5G NR

In 5G NR, the gNB (base station) performs uplink detection on the PUSCH (Physical Uplink Shared Channel). With up to 256 antenna ports in FR1 and massive MIMO in TDD mode, the detection problem is precisely the one formulated here. The standard leaves the detection algorithm to the vendor implementation — MRC for initial access, MMSE or MMSE-SIC for data traffic — making this chapter directly relevant to 5G receiver design.

Quick Check

In the uplink model $\mathbf{y} = \mathbf{H}\mathbf{x} + \mathbf{w}$ , what is the dimension of the received vector $\mathbf{y}$ if the BS has 64 antennas and there are 8 users?

$8 \times 1$

$64 \times 1$

$64 \times 8$

$8 \times 8$

Correction:

64 \times 1

The received vector has one entry per BS antenna: $\mathbf{y} \in \mathbb{C}^{N_t}$ with $N_t = 64$ .

Uplink System Model