Chapter Summary

Chapter Summary

Key Points

• 1.

  Testbed landscape. The massive-MIMO testbed community has converged on a small set of reference platforms — LuMaMi (Lund), ArgosV3 (Rice), HHI testbeds (Fraunhofer Berlin), OAI, srsRAN, and the CommIT spin-off Massive Beams. Each combines a coherent array of $N_t$ antennas, a real-time baseband, and an instrumentation layer, and each reports a measurable reality gap between theoretical capacity and measured throughput.

• 2.

  Real-time compute budget. 5G NR at 30 kHz numerology fixes $T_{\rm slot} = 500~\mu$s, leaving $\sim 300~\mu$s of compute window after pilots and TDD guard. Zero-forcing scales as $\Theta(N_tK^{2} + K^{3})$ complex multiplies per slot, dominated by the Gram-matrix step in the massive-MIMO regime. The FPGA, SoC, and x86 architectures each trade latency, area, and developer productivity to fit that budget.

• 3.

  Fixed-point precision. Under a $b$-bit mantissa the effective SINR obeys $\gamma_{\rm fxp}(b) \approx \gamma_0 / (1 + \gamma_0 c_{\rm arch} 2^{-2b})$, where $c_{\rm arch} \sim N_tK$. A LuMaMi-class testbed needs $b \gtrsim 10$ to stay within a 0.3 dB penalty of floating-point, with extra bits reserved for accumulator headroom.

• 4.

  Reciprocity calibration. The TDD reciprocity claim holds only after correcting for RF-chain mismatch. Rogalin-Caire (2014) proved that relative calibration of $\mathbf{C} = \mathbf{D}_{\rm rx}^{-1}\mathbf{D}_{\rm tx}$ is sufficient, recoverable by the BS-internal Argos protocol, and self-tracking against temperature drift at a sub-second cadence in outdoor conditions.

• 5.

  Synchronization. Frame-level sync from PSS/SSS, symbol alignment within the cyclic prefix, and carrier frequency offset well below $\Delta f$ are all required simultaneously. The massive-MIMO array amplifies the ICI induced by residual CFO rather than averaging it out, yielding a per-array SINR ceiling. Phase noise becomes dominant at FR2 because it scales as $f_0^{2}$. Cell-free testbeds use GPSDO plus scheduled inter-AP calibration to reach coherent joint transmission.

• 6.

  Over-the-air measurement campaigns. Channel sounders and testbeds produce the raw data that feeds 3GPP TR 38.901 channel models and the Rel-18 AI/ML workitem. The reality gap between theoretical and measured rates decomposes into systematic modeling bias, hardware residuals, and non-Gaussian pathologies (blockage, mobility spikes). Curated datasets with outlier annotations and non-random train/test splits are the professional standard for learned physical-layer components.

• 7.

  CommIT contributions. The Rogalin-Bursalioglu-Papadopoulos- Caire (2014) relative-calibration theorem underpins every modern TDD massive MIMO testbed; the Gottsch-Ito-Caire (2023) distributed-real-time analysis formalizes the compute budget for cell-free; and the TU Berlin spin-off Massive Beams is the commercial embodiment of the cell-free ideas developed in Chapters 11--15.