Ferkans — Interactive Telecom Tutor

DoF Is Only Half the Story

DoF analysis captures the high-SNR slope of the sum-rate but ignores the intercept — the additive constant that determines rate at operationally relevant SNRs (5–30 dB for most deployments). A scheme with $\mathrm{DoF} = t + L$ may underperform at finite SNR if its pre-log constant is small (weak effective SNR) or if it incurs large processing overhead.

This section analyzes the finite-SNR performance of the Lampiris- Caire scheme: how the sum-rate depends on SNR, how it compares to pure MIMO and pure caching, and where the practical break-even point lies. The analysis informs deployment: in what SNR regime does multi-antenna coded caching pay off?

Theorem: Finite-SNR Sum-Rate Approximation

For the Lampiris-Caire scheme with perfect CSIT, integer $t$ , $L$ antennas, $K$ users, the achievable sum-rate at transmit SNR $\rho$ is $R_{\mathrm{sum}}(\rho) \;=\; (t + L) \log_2(1 + \beta \rho) + O(1),$ where $\beta$ is a scheme-dependent constant (0 < $\beta < 1$ ) capturing the ZF interference-leakage penalty.

The DoF is the slope in $\log_2 \rho$ ; the constant $\beta$ accounts for the beamforming loss. For well-conditioned channels, $\beta \to 1$ ; for poorly conditioned (e.g., collinear users), $\beta$ can be small.

Proof

Per-beam SINR

Each coded-XOR stream beamed by zero-forcing has SINR $\rho \beta_k$ for user $k$ , where $\beta_k$ depends on the channel conditioning of the $(t+L)$ -group.

Per-beam rate

$R_{\text{beam},k} = \log_2(1 + \rho \beta_k)$ . For equal-power allocation, $\mathbb{E}_k[\beta_k] \approx 1 - L/K$ in the large- $K$ limit (no interference leakage in ZF).

Sum-rate

Sum over all $(t+L)$ -groups and $L$ streams per group, divided by delivery length: $R_{\text{sum}} \;=\; (t + L) \log_2(1 + \beta \rho) + O(1).$ The $O(1)$ hides channel-dependent constants and the MAN bookkeeping overhead.

Sum-Rate vs SNR: Cache + MIMO vs Alternatives

Compare the sum-rate of the Lampiris-Caire scheme (blue, DoF = $t+L$ ) with pure MIMO BC (red dashed, DoF = $L$ ) and pure caching (green dotted, DoF = $t+1$ , single antenna). Observe (1) the asymptotic slope difference — DoF matters at high SNR; and (2) the crossover with pure MIMO at low SNR — caching adds modest gain until beamforming gets expensive.

Parameters

Users K10

Antennas L4

Memory ratio M/N0.2

Example: Crossover SNR for Cache Benefit

For $K = 10$ , $L = 4$ , $M = 2$ , $N = 10$ , find the SNR at which the Lampiris-Caire scheme's sum-rate exceeds pure MIMO by 10% (assume $\beta = 0.8$ ).

Solution

DoF

$t = 2$ , $L = 4$ , so $\mathrm{DoF} = t + L = 6$ (Lampiris-Caire) vs $L = 4$ (pure MIMO).

Sum-rate formulas

$R_{\text{LC}} \approx 6 \log_2(1 + 0.8 \rho)$ . $R_{\text{MIMO}} \approx 4 \log_2(1 + 0.8 \rho)$ .

Crossover

Ratio $R_{\text{LC}}/R_{\text{MIMO}} = 6/4 = 1.5$ . This holds for any SNR where both schemes are well above threshold (say $\rho > 1$ or 0 dB). The LC scheme yields 50% more rate at all reasonable SNRs. The 10% threshold is met effectively from 0 dB onward.

Operational takeaway

The DoF advantage translates into rate at any SNR where the scheme operates efficiently. The caveat: at very low SNR (noise-limited, < -10 dB), the beamforming overhead can make pure MIMO worse than no-beamforming schemes.

⚠️Engineering Note

Deployment Constraints in 5G NR

Translating Lampiris-Caire to a deployed 5G NR system faces several practical constraints:

MBSFN (Multi-Broadcast Single-Frequency Network). 5G multicast features enable shared-content delivery to groups of users. A cache-aided extension has been prototyped but is not widely deployed.
CSIT acquisition. For the $L = 4$ or $L = 8$ systems typical in sub-6 GHz, CSIT acquisition is feasible via SRS / UL pilots. At mmWave (L = 64+), CSIT is a bottleneck, and coded caching's CSI-independent gain becomes more attractive.
MCS assignment. The per-user SNR post-ZF may vary; MCS should be chosen conservatively.
Latency. The delivery phase combines several $(t+L)$ -groups serially; latency per delivery round is $O(\binom{K}{t+L}/L)$ slots. For latency-sensitive services, coded caching is not a good fit.
Subpacketization. Already constrained to $O(10^8)$ bytes per file; multi-antenna coded caching adds a factor of $\binom{K}{t+L-1}/\binom{K}{t}$ on top. Chapter 14's PDAs extend to multi-antenna settings.

The research-to-practice gap for multi-antenna coded caching remains significant. Recent CommIT papers (Lampiris-Bhattacharjee- Caire 2020) address realistic CSIT and fading; Piepenbrink-Caire 2024 considers practical PHY prototypes.

Practical Constraints

•
5G NR PDCP supports bundled multicast for broadcast content
•
Practical L = 4-8 (sub-6) or 16-64 (mmWave)
•
CSIT fidelity bounds DoF beyond t+L in interference-limited regimes
•
Subpacketization factor $\binom{K}{t+L-1}/\binom{K}{t}$

Coded Caching with No CSIT

A striking follow-up result: coded caching can partially substitute for CSIT. Without CSIT, the pure-MIMO DoF collapses from $L$ to $1$ (single-user time-sharing). But cache-aided delivery can still achieve $\mathrm{DoF} = t + 1$ (the single-antenna MAN result), without requiring CSIT — the cached side information lets the transmitter exploit XOR multicasting blindly.

More refined results (Lampiris-Elia-Caire 2018) show that when CSIT is partially available (say, only user-index information, not fine-grained channel quality), the DoF interpolates between $t+1$ (no CSIT) and $t+L$ (full CSIT). The practical upshot is that coded caching is a CSIT-economical design — it extracts DoF from the cache budget without demanding high-quality channel information.

Common Mistake: DoF Analysis Assumes Perfect CSIT

Mistake:

Quoting $\mathrm{DoF} = t + L$ as the achievable rate in a deployed system without qualifying the CSIT assumption.

Correction:

The DoF $t+L$ requires perfect instantaneous channel state information at the transmitter. In realistic systems:

No CSIT: $\mathrm{DoF} = t + 1$ (cache gain survives, spatial gain is lost).
Delayed CSIT: $\mathrm{DoF}$ between $t+1$ and $t+L$ depending on the delay-to-coherence ratio.
Quantized CSIT ( $b$ feedback bits): scaling analysis shows $\mathrm{DoF}(b) = t + L(1 - 2^{-b/L})$ approximately.

These caveats are important when comparing theoretical and deployed system performance.

Finite-SNR Analysis and Practical Tradeoffs