Massive MIMO SINR Scaling Explained
Understanding the appearance of \(1/\sqrt{M}\) in imperfect CSI scenarios
Introduction
In Massive MIMO systems, understanding how the signal-to-interference-plus-noise ratio (SINR) scales with the number of base-station antennas \(M\) is crucial. This article presents a clear, simplified derivation to explain why \(1/\sqrt{M}\) scaling appears in the case of imperfect channel state information (CSI).
System Setup & Assumptions
- Single desired user (no inter-user interference).
- Uplink transmit power per user = \(p_u\), noise variance per antenna = 1.
- Base station has \(M\) antennas; channel vector \(g\).
- Channel estimated via uplink pilots. Pilot power scales with data power \(p_u\).
- Matched filter combiner: \(v = \hat{h}\), where \(\hat{h}\) is the estimated channel.
- We consider dominant scalings with \(M\) (big-\(M\) asymptotics), ignoring multiplicative constants.
Scaling of Key Quantities
The dominant scalings often used in literature are:
- Estimated channel norm: \(\|\hat{h}\| \propto p_u M\)
- Estimation error contribution: \(\propto p_u M\)
- Noise power after combining: \(\propto \|\hat{h}\|^2 \propto p_u^2 M^2\)
Matched-Filter Output — Dominant Scalings
After matched filtering, the simplified dominant scalings are:
- Signal power: \(\propto p_u^2 M^2\)
- Interference + noise: \(\propto p_u M + M\)
Hence, the simplified SINR scales as:
$$SINR \approx \frac{p_u^2 M^2}{p_u M + M}$$
Finding the Power Scaling to Keep SINR Constant
Let transmit power scale as \(p_u = c M^{-\alpha}\). Then:
$$
SINR \propto \frac{c^2 M^{2 - 2\alpha}}{c M^{1 - \alpha} + M} \\
= c^2 \frac{M^{1 - 2\alpha}}{c M^{-\alpha} + 1}
$$
For large \(M\), \(c M^{-\alpha} \to 0\), so:
$$SINR \propto c^2 M^{1 - 2\alpha}$$
To keep SINR constant, set the exponent to zero:
$$1 - 2\alpha = 0 \quad \to \quad \alpha = 1/2$$
Therefore, transmit power should scale as:
$$p_u = \frac{c}{\sqrt{M}}$$
Quick Check: Perfect CSI Case
With perfect CSI (no estimation error), the SINR scales differently:
$$SINR \propto \frac{p_u^2 M^2}{M} = p_u^2 M$$
Here, the transmit power can be reduced faster than \(1/\sqrt{M}\) while maintaining SINR.
Here, the denominator is only 'M' because
- In the case of MIMO with perfect Channel State Information (CSI), interference is effectively nullified (0 + M), as all interference is canceled due to signal orthogonality.
Intuition
Imperfect CSI couples the combiner quality with pilot energy (tied to \(p_u\)). This coupling introduces a term \(\propto M\) in the denominator, which algebraically forces the \(p_u \propto 1/\sqrt{M}\) scaling. Perfect CSI allows stronger power reduction because the denominator does not include the estimation error term.
Conclusion
The \(1/\sqrt{M}\) scaling in massive MIMO with imperfect CSI arises from balancing the power of the desired signal and the amplified noise/error contributions. This simple derivation captures the essential algebra behind the scaling laws frequently cited in literature.