Ferkans — Interactive Telecom Tutor

Models Meet Reality: Channel Measurement Campaigns

All channel models in this chapter are validated against empirical measurements. Without measurements, models are untethered speculation. This section surveys what wideband MIMO channel measurements actually reveal at sub-6 GHz and mmWave — the two frequency regimes that define current 5G and near-future 6G deployments. The key message: the channel is fundamentally different at mmWave, and the design principles must adapt accordingly.

Sub-6 GHz vs. mmWave: Propagation Comparison

Property	Sub-6 GHz (e.g., 3.5 GHz)	mmWave (e.g., 28 GHz)
Wavelength	$\approx 8.6$ cm	$\approx 1.1$ cm
Path loss exponent (NLOS)	$3.0$ – $3.8$	$3.5$ – $4.5$
Coverage range	Up to 2–5 km (macro)	100–500 m (micro)
RMS delay spread	$10$ – $200$ ns	$1$ – $50$ ns
Angular spread (ASD)	$5°$ – $30°$	$1°$ – $10°$ (sparse)
Number of clusters $L$	$3$ – $10$	$1$ – $3$
Blockage sensitivity	Low (diffraction)	High (no diffraction)
Effective MIMO rank	Moderate ( $3$ – $10$ )	Low ( $1$ – $3$ )
Coherence bandwidth	$100$ kHz– $2$ MHz	$1$ – $20$ MHz
Doppler spread (60 km/h)	$\approx 195$ Hz	$\approx 1.6$ kHz
Penetration loss (concrete)	$5$ – $10$ dB	$30$ – $50$ dB
Dominant channel model	Correlated Rayleigh (Kronecker/one-ring)	Sparse geometric (few paths)

Definition:
Channel Sounding: Basic Measurement Methodology

A channel sounder measures the wideband channel impulse response $h(t, \tau)$ by transmitting a known wideband probe signal and correlating the received signal. For MIMO channel sounding with $N_t$ transmit and $N_r$ receive antennas, sequential switching between antenna pairs measures the full $N_r \times N_t$ matrix in approximately $N_t \times N_r \times T_{\text{switch}}$ seconds.

Key measurement system parameters:

Bandwidth $B$ : determines delay resolution $\Delta\tau = 1/B$
Measurement repetition rate $f_s$ : must satisfy $f_s > 2 f_{D,\max}$ (Nyquist in time)
Array aperture $D = (N-1)d$ : determines angular resolution $\approx \lambda/D$
Dynamic range: needed path loss range (typically 80–110 dB)

What Sub-6 GHz Measurements Reveal

Key findings from large-scale sub-6 GHz massive MIMO measurement campaigns (Lund University, Bristol, Aalborg, TU Berlin, NYU):

1. Channel hardening and favorable propagation hold, but imperfectly: For $N_t = 128$ and 12 single-antenna UEs, the measured effective SNR spread (channel hardening quality) was within 3 dB of the theoretical i.i.d. prediction. Favorable propagation held to within 6 dB of perfect orthogonality.

2. Spatial correlation is significant: Typical measured correlation coefficient between adjacent antenna elements at $d = \lambda/2$ spacing: $|[\mathbf{R}_t]_{1,2}|/[\mathbf{R}_t]_{1,1} \approx 0.3$ – $0.7$ in outdoor macro environments at 2.6 GHz.

3. Non-stationarity across large arrays: For arrays longer than 1 m (approximately 12 elements at 2.6 GHz), the channel statistics change measurably along the array — a precursor to the near-field effects in XL-MIMO (Chapter 17).

4. Capacity is model-sensitive: Using i.i.d. Rayleigh to predict sum-rate overestimates actual capacity by 15–40% in outdoor macro scenarios, depending on the angular spread distribution.

,

What mmWave Measurements Reveal

Key findings from mmWave channel measurement campaigns at 28 GHz, 39 GHz, 60 GHz, and 73 GHz (NYU Wireless, Samsung, Qualcomm, Ericsson, NTT DOCOMO):

1. Extreme sparsity: The number of significant propagation clusters is $L \approx 1$ – $3$ in outdoor environments. The virtual channel $\tilde{\mathbf{H}}$ often has fewer than 5 nonzero entries for a $64 \times 4$ MIMO channel — meaning compressed sensing can recover the channel from as few as 10–20 pilots.

2. Severe blockage: The human body causes 15–30 dB attenuation. Vehicle blockage at 28 GHz causes link outages lasting 50–200 ms. This makes beam management a critical protocol function in 5G NR FR2 (Chapter 22).

3. Oxygen absorption at 60 GHz: At 60.0 GHz, atmospheric oxygen resonance causes $\approx 15$ dB/km additional attenuation — limiting outdoor range to $\leq 200$ m. This makes 60 GHz suitable only for backhaul or very dense indoor deployments.

4. Large arrays are not strongly correlated (on a per-element basis): Despite fewer paths, the small wavelength at mmWave means the array must be physically large (many elements) to achieve a given aperture. The correlation between elements at $d = \lambda/2$ is similar to sub-6 GHz, but fewer effective channels coexist.

,

🚨Critical Engineering Note

Coherence Time and Channel Aging in Massive MIMO

The coherence time $T_c \approx 1/(2f_D)$ constrains how long a channel estimate remains valid. For a UE moving at speed $v$ :

Sub-6 GHz at $f_0 = 3.5$ GHz, $v = 30$ km/h: $f_D = v f_0/c \approx 97$ Hz, $T_c \approx 5$ ms.
mmWave at $f_0 = 28$ GHz, $v = 30$ km/h: $f_D \approx 778$ Hz, $T_c \approx 0.6$ ms.

For a 5G NR subframe of 1 ms with $N_t = 64$ pilots:

Sub-6 GHz: Pilot overhead $\approx 64/(T_c \cdot B_c) = 64/(5000) \approx 1.3\%$ (manageable).
mmWave: Pilot overhead $\approx 64/(600) \approx 10.7\%$ — significant.

This analysis explains why mmWave systems use beam-level tracking (slowly varying large-scale structure) rather than element-level tracking, and why angular-domain sparsity (few paths → few pilots needed) is critical for mmWave MIMO efficiency.

Practical Constraints

•
5G NR FR1 (sub-6 GHz): coherence time $> 1$ ms at pedestrian speeds — standard TDD reciprocity works well
•
5G NR FR2 (mmWave): $T_c < 1$ ms at vehicular speeds — beam sweeping protocols essential
•
3GPP requires beam management latency $< 10$ ms for UE mobility (TS 38.300)

📋 Ref: 3GPP TR 38.901, Section 7.6 (Doppler); 3GPP TS 38.300, Section 9.3

Why This Matters: From Measurement to Massive MIMO Design Rules

The measurement findings above translate directly into massive MIMO design rules:

Sub-6 GHz outdoor macro: Use one-ring or Kronecker covariance with ASD $\approx$ 5–15°. Design precoding (JSDM) around the dominant eigenvectors of $\mathbf{R}_t$ . Channel estimation benefits from covariance-based MMSE. Pilot contamination is a real problem — use covariance-based decontamination.
mmWave: Use sparse geometric model with $L \leq 3$ paths. Compressed sensing estimation (Chapter 3) is feasible and necessary. Hybrid beamforming (Chapter 20) with $N_{\text{RF}} \approx L$ RF chains is near-optimal. Near-field effects become significant for arrays larger than 5 cm × 5 cm.
Indoor: Use near-i.i.d. model (high angular spread). Pilot contamination is less severe but path loss limits range. Dense deployment of small arrays preferred.

Sub-6 GHz vs mmWave Propagation Geometry — Illustration of the fundamental difference in propagation at sub-6 GHz and mmWave. At sub-6 GHz (left), diffraction enables non-line-of-sight coverage via multiple scattering paths with moderate angular spread. At mmWave (right), propagation is quasi-optical with 1–3 dominant reflected/scattered paths and high blockage sensitivity.

Common Mistake: Massive MIMO at mmWave Is Hybrid, Not Digital

Mistake:

Assuming that massive MIMO (large $N_t$ ) at mmWave uses fully digital processing with one RF chain per antenna, as in sub-6 GHz.

Correction:

At mmWave, the hardware cost and power consumption of fully digital architectures is prohibitive. A 64-antenna fully digital mmWave system at 28 GHz would require 64 ADC/DAC pairs at $\geq 2$ GHz sampling rate, consuming $>10$ W in ADCs alone (vs $<0.5$ W for the same at 3.5 GHz). In practice, mmWave massive MIMO uses hybrid beamforming with $N_{\text{RF}} \approx 4$ –16 RF chains and $N_t = 32$ –256 antenna elements connected via phase shifters. The spatial multiplexing gain is limited to $N_{\text{RF}}$ streams, not $\min(N_t, N_r)$ .

The sparse channel (few paths) aligns with this: with $L \leq 3$ paths, $N_{\text{RF}} = 4$ RF chains are already sufficient to capture nearly all channel capacity.

Key Takeaway

Channel measurements are the ground truth. Sub-6 GHz outdoor channels are spatially correlated (ASD 5–15°, effective rank 3–10), confirming that the one-ring model is a reasonable approximation and that JSDM is practically motivated. mmWave channels are extremely sparse ( $L \leq 3$ paths), making hybrid beamforming with a few RF chains near-optimal and compressed sensing channel estimation essential. The 3GPP TR 38.901 model captures both regimes with scenario-specific parameters.

Quick Check

At 28 GHz in an outdoor urban environment, a measured channel has $L = 2$ dominant paths with angles of departure at $\psi_1 = -0.2$ and $\psi_2 = 0.1$ (in spatial frequency). For a $N_t = 64$ ULA, approximately how many nonzero entries does the virtual channel $\tilde{\mathbf{H}}$ have (per receive antenna)?

64

2 (approximately)

32

$\log_2(64) = 6$