5G NR is built on a scalable numerology indexed by \(\mu \in \{0,1,2,3,4\}\). Every time and frequency quantity derives from two atomic time units defined in TS 38.211 §4.1.

Numerology Table — all \(\mu\) values TS 38.211 Table 4.3.1-1

Frame Hierarchy (\(\mu = 1\))

Normal CP Lengths at \(f_s = 122.88\) MHz, \(\mu = 1\) TS 38.104 Table 5.3.2-1

The OFDM symbol duration (excluding CP) is \(N_{\text{FFT}} / f_s = 4096 / 122.88\ \text{MHz} = 33.33\ \mu\text{s}\). Two CP lengths apply per slot:

Total slot duration check: \(2 \times (4096+512) + 12 \times (4096+288) = 11216 + 52608 = 63488\) samples \(= 63488 / 122.88 \times 10^6 = 0.5168\ \text{ms} \approx 0.5\ \text{ms}\). ✓

NR DL uses CP-OFDM. The baseband time-domain signal for OFDM symbol \(l\) is formed by an inverse DFT (IFFT) of the frequency-domain complex coefficients followed by insertion of a cyclic prefix.

Time-Domain Signal Model

IFFT Mapping: Active Subcarriers

The 4096-point IFFT is populated as follows:

The remaining \(4096 - 1584 = 2512\) FFT bins are set to zero (guard subcarriers + DC null), preventing spectral leakage into adjacent channels. The sampling clock \(f_s = 122.88\ \text{MHz}\) satisfies the Nyquist criterion for the full FFT bandwidth \(N_{\text{FFT}} \times \Delta f = 4096 \times 30\ \text{kHz} = 122.88\ \text{MHz}\).

Cyclic Prefix — Multipath Robustness

Windowing TS 38.211 §5.4.1

To reduce out-of-band emissions, a raised-cosine (RC) window is applied at symbol boundaries. The window overlaps adjacent CP regions by \(N_W\) samples, tapering transitions smoothly rather than applying a hard rectangular cut-off:

The NR resource grid is a two-dimensional time–frequency lattice. Every cell in the grid is a Resource Element (RE); groups of 12 adjacent subcarriers form a Physical Resource Block (PRB).

Resource Element & PRB TS 38.211 §4.4.1

Resource Grid Dimensions — Our System

RE Mapping: Channel Types

Within each slot, REs are assigned to different physical channels and signals:

VRB → PRB Mapping TS 38.211 §4.4.4

The scheduler operates on Virtual Resource Blocks (VRBs) — a contiguous numbered set used for DCI signalling. VRBs are mapped to Physical Resource Blocks (PRBs) via one of two modes:

• Non-interleaved: VRB \(n\) maps directly to PRB \(n + N_{\text{start}}\). Simple; frequency-selective scheduling; no frequency diversity.
• Interleaved: VRBs permuted across the bandwidth using a block interleaver (bundle size 2 or 4 PRBs). Provides frequency diversity at the cost of reduced frequency-selective gain — useful for URLLC or cell-edge UEs.

Time Division Duplex (TDD) shares a single carrier between DL and UL by time-multiplexing directions across slots. Our deployment uses a 9-slot repeating pattern over \(9 \times 0.5\ \text{ms} = 4.5\ \text{ms}\).

DDDSUUDDD Visual — 9-slot pattern (4.5 ms period at μ=1)

Each slot = 0.5 ms. DL full downlink   S special: 10DL+2GP+2UL   UL full uplink   Total period = 9 × 0.5 ms = 4.5 ms

DDDSUUDDD Slot Map (\(\mu=1\), 9-slot period) TS 38.211 Table 4.3.2-1

TDD Slot Timeline (4.5 ms period)

DL = blue, Special = purple, UL = green. Total period = 4.5 ms. Pattern repeats every 9 slots (not aligned to subframe boundary at μ=1).

Special Slot (S3) — Symbol-Level Breakdown

Special slot format 10 DL / 2 GP / 2 UL:

Guard Period (GP) — Propagation Delay Budget

Effective DL/UL Duty Cycle

14.1 — 32-TRX Antenna Array

The system has 32 TRX — 32 independent RF chains at the gNB, each comprising a PA, LNA, and ADC/DAC pair. Each TRX maps to one logical antenna port in the PDSCH precoder. Two dominant physical layouts are used in practice:

The O-RAN split 7-2x interface carries the 32 antenna port I/Q streams between the DU and RU. Per O-RAN WG4 spec, the DU computes the precoding weights and passes pre-beamformed I/Q (Category B) or raw per-port I/Q (Category A) — 32 TRX Category A = 32 separate streams over the fronthaul.

14.2 — DL MIMO Channel Model

The received signal model for a single UE with \(N_r\) receive antennas:

• \(\mathbf{H} \in \mathbb{C}^{N_r \times 32}\) — channel matrix (UE receive × gNB transmit ports)
• \(\mathbf{W} \in \mathbb{C}^{32 \times \nu}\) — precoding matrix (\(\nu\) = transmission rank)
• \(\mathbf{x} \in \mathbb{C}^{\nu \times 1}\) — transmitted data vector across \(\nu\) layers
• \(\mathbf{n} \in \mathbb{C}^{N_r \times 1}\) — AWGN noise: \(\mathbf{n} \sim \mathcal{CN}(0, N_0 \mathbf{I})\)

The effective channel seen after precoding is \(\mathbf{H}\mathbf{W} \in \mathbb{C}^{N_r \times \nu}\). SVD decomposition gives \(\mathbf{H} = \mathbf{U}\boldsymbol{\Sigma}\mathbf{V}^H\), and the optimal precoder for SU-MIMO is \(\mathbf{W}^* = \mathbf{V}_{:,1:\nu}\) (first \(\nu\) right singular vectors), matched to the strongest channel eigenmodes.

14.3 — Type I Single-Panel Codebook (TS 38.214 §5.2.2.2)

The 5G NR codebook uses a two-stage structure:

• W₁ — wideband beam group selection: selects a group of oversampled DFT beam vectors. One W₁ per PMI reporting instance (wideband, changes slowly with large-scale channel). \(\mathbf{W}_1 \in \mathbb{C}^{N_t \times 2L}\), where \(L\) is the number of candidate beams (typically \(L=2\) or \(L=4\)).
• W₂ — subband co-phasing and beam selection: selects beam(s) within the group and applies inter-polarization co-phasing. One W₂ per PRG (Physical Resource Group, subband granularity). \(\mathbf{W}_2 \in \mathbb{C}^{2L \times \nu}\).

For our N1=8, N2=2, O1=4, O2=4 UPA (8 columns, 2 rows, 4× oversampling each):

• Horizontal DFT beams: \(N_1 \times O_1 = 8 \times 4 = 32\) beams
• Vertical DFT beams: \(N_2 \times O_2 = 2 \times 4 = 8\) beams
• Total beam pairs (one polarization): \(32 \times 8 = 256\)
• Dual-polarization ports: total codebook spans \(256\) dual-pol beam directions

14.4 — Type II Enhanced Codebook (TS 38.214 §5.2.2.3)

Type I codebook is single-beam selection per polarization. Type II provides a linear combination of multiple beams:

where \(\mathbf{B}\) is a matrix of \(L=2\) or \(L=4\) DFT beam vectors per polarization, and \(\mathbf{C}\) contains amplitude + phase coefficients for each beam–layer combination. The UE reports these coefficients in the PMI — significantly higher feedback overhead but captures the true channel subspace far better than a single beam.

Note on Type II max layers: Type II SP and Port Selection codebooks (TS 38.214 §5.2.2.3.1) support at most 2 layers per UE (ν ∈ {1, 2}). Enhanced Type II (Rel-16) extends this to ν ≤ 4. None of the Type II variants support 16 layers for a single UE — "16 layers" in this system is a MU-MIMO sum rank achieved by scheduling 8 UEs × rank-2 each.

14.5 — Achieving 16-Layer DL with 32 TRX

Interpretation 1 — SU-MIMO up to 8 layers:

• Single UE receives 8 independent PDSCH layers simultaneously
• Requires UE with 8 receive antennas (uncommon for handsets; used in fixed wireless access)
• Uses Type I SP or Type II codebook with ν = 8
• Peak UE throughput ≈ 8 × (single-layer rate) under good channel conditions

Interpretation 2 — MU-MIMO sum rank 16 (most practical):

• 8 UEs scheduled simultaneously on the same time-frequency resources
• Each UE receives rank-2 PDSCH → 2 layers/UE
• Total sum rank = 8 × 2 = 16 spatial streams
• gNB computes joint precoder W such that inter-user interference is suppressed
• This is the dominant mode in commercial massive MIMO deployments

14.6 — MU-MIMO Precoding Design

For MU-MIMO with \(K\) simultaneously served UEs, the aggregate channel matrix is:

The received signal at UE \(k\) with MU precoding:

Zero-Forcing (ZF) precoder — eliminates IUI exactly (requires perfect CSI):

MMSE precoder — balances IUI suppression vs. noise amplification:

At high SNR, MMSE → ZF. At low SNR, MMSE → matched filter (maximum ratio transmission). The per-UE columns of \(\mathbf{W}\) are then normalized to satisfy total transmit power \(P_T\).

14.7 — Channel Estimation via SRS (TDD Reciprocity)

In TDD mode, the UL and DL channels are reciprocally related: \(\mathbf{H}_{DL} \approx \mathbf{H}_{UL}^T\) (after RF calibration). UEs transmit Sounding Reference Signals (SRS) in UL slots; the gNB estimates \(\mathbf{H}_{UL}\) from all 32 receive ports simultaneously and uses \(\mathbf{H}_{DL} = \mathbf{H}_{UL}^T\) to compute the precoder.

• SRS bandwidth: configurable from 4 PRBs to full 132 PRBs
• SRS comb: COMB-2 (every other subcarrier) or COMB-4 for UE multiplexing
• With DDDSUUDDD: UL slots at positions 4,5 → SRS available every 0.5–1 ms
• Channel coherence time: \(T_{coh} \approx \frac{0.423}{f_D}\), where \(f_D = \frac{v \cdot f_c}{c}\)

At \(v = 60\) km/h = 16.7 m/s and \(f_c = 3.5\) GHz:

At 2.2 ms coherence time, the precoder computed from SRS in slot \(n\) remains valid for approximately 4–5 slots (each slot = 0.5 ms with μ=1). The gNB must refresh the precoder at least every 4 slots for pedestrian/vehicular UEs.

14.8 — Beamforming Gain with 32 TRX

For a uniform linear array of \(N_t = 32\) elements, the array factor in direction \(\theta\) is:

With matched-filter (MRT) weights \(w_n = e^{-j n k d \sin\theta_0}\) steered to \(\theta_0\):

• Peak beamforming gain: \(G_{bf} = 10\log_{10}(32) \approx 15.1\) dB over isotropic transmit
• HPBW (half-power beamwidth) for \(d=\lambda/2\): \(\text{HPBW} \approx 0.886 \times \frac{2}{N_t} \times \frac{180°}{\pi} \approx \frac{102°}{32} \approx 3.2°\)
• First sidelobe: −13.3 dB below main lobe (Sinc-pattern for uniform weights)
• With tapering (Chebyshev/Taylor window): sidelobe → −30 dB at cost of 20–30% HPBW widening

14.9 — SU-MIMO Capacity (Waterfilling)

With SVD decomposition \(\mathbf{H} = \mathbf{U}\boldsymbol{\Sigma}\mathbf{V}^H\), the MIMO channel decomposes into \(\nu\) parallel scalar channels with gains \(\sigma_i^2\). Waterfilling power allocation maximizes capacity:

\(\mu\) is the water level chosen to satisfy \(\sum_i P_i = P_T\). In practice, 5G NR uses equal power per layer (no waterfilling in standardized operation), with MCS adaptation per layer via separate CQI/RI feedback.

14.10 — Beam Pattern: 32-Element ULA Steering (Plot)

Normalized array factor \(|A(\theta)|^2\) (dB) for a 32-element ULA (\(d=\lambda/2\)) steered to five angles: 0°, ±15°, ±30°. Main lobe HPBW ≈ 3.2°; first sidelobe ≈ −13 dB.

14.11 — Capacity vs MU-MIMO Rank (Plot)

Sum spectral efficiency (bps/Hz) vs. number of simultaneously served UEs for 32-TRX MU-MIMO at three SNR operating points. Capacity plateaus at high rank due to inter-user interference with imperfect ZF precoding (10% residual IUI assumed at high rank).

14.12 — PMI Feedback Structure for 32-Port Array

The CSI feedback chain for a UE with 32-port CSI-RS:

1. gNB transmits CSI-RS on up to 32 ports (NZP-CSI-RS, periodically or aperiodically)
2. UE measures CSI-RS, estimates the effective channel \(\hat{\mathbf{H}}\)
3. UE selects optimal RI (rank indicator), PMI (W1 beam group + W2 co-phasing), CQI (MCS index)
4. UE reports RI/PMI/CQI via PUCCH (periodic) or PUSCH (aperiodic, higher precision)
5. gNB scheduler uses reported PMI to select precoding matrix W for next PDSCH transmission

With PRG (Physical Resource Group) size \(P = 8\) PRBs, the 132-PRB band is split into \(\lceil 132/8 \rceil = 17\) subbands. The UE reports one wideband W1 + 17 subband W2 values per CSI report — capturing frequency-selective precoding with fine granularity.

15.1 — Scheduler Architecture (128 UEs, 16 Spatial Layers)

The MAC scheduler in the gNB DU runs every slot (0.5 ms for μ=1), deciding:

• Which subset of the 128 connected UEs to schedule in the current slot
• How many PRBs and which frequency resources to assign each UE
• Which MCS (modulation + code rate) to use per UE (target BLER = 10%)
• What MIMO rank and precoding to apply (up to 16 sum spatial layers across all UEs)
• Which HARQ process to use and whether this is initial transmission or retransmission

15.2 — Proportional Fair (PF) Scheduling

The Proportional Fair criterion balances throughput maximization with long-term fairness. For UE \(k\) at slot \(t\):

where:

• \(R_k(t)\) = instantaneous achievable rate (from CQI → MCS lookup) in current slot
• \(\bar{R}_k(t)\) = exponentially weighted moving average (EWMA) of past throughput:

Typical window \(T_c = 100\) slots (50 ms). The PF metric \(R_k / \bar{R}_k\) is:

• > 1: UE \(k\) is experiencing better-than-average channel → prefer scheduling now
• < 1: UE \(k\) has been receiving good throughput → deprioritize, give others a chance
• = 1: UE at its average → neutral

15.3 — Spatial Compatibility Check for MU-MIMO Pairing

For MU-MIMO to work, the selected UEs must have sufficiently orthogonal channels — otherwise the ZF precoder will noise-amplify rather than interference-cancel. The pairing criterion:

where \(\hat{\mathbf{h}}_k = \mathbf{h}_k / \|\mathbf{h}_k\|\) is the normalized channel direction. Typical threshold \(\gamma_{th} = 0.1\) (−10 dB channel orthogonality).

In Urban Macro (UMa) at 3.5 GHz, UEs separated by > 20° angular spread typically achieve \(|\hat{\mathbf{h}}_i^H \hat{\mathbf{h}}_j|^2 \leq -15\;\text{dB}\) — sufficient for MU pairing with 8 layers and ZF precoding. The gNB groups UEs by SRS-estimated spatial direction and selects the highest-PF-metric subset with pairwise orthogonality \(\leq \gamma_{th}\).

15.4 — Resource Allocation: VRB → PRB Mapping

The scheduler assigns Virtual Resource Blocks (VRBs); the lower layer maps them to Physical Resource Blocks (PRBs):

• Non-interleaved mapping (default): VRB \(n\) → PRB \(n\). Simple, predictable, used for most scheduled PDSCH (DCI format 1_0 and 1_1).
• Interleaved mapping: VRBs permuted across the bandwidth via a block interleaver → frequency diversity. Used for SIB1, RA-RNTI, paging transmissions.

DCI format 1_1 resource allocation types:

• Type 0: RBG (Resource Block Group) bitmap. For N_RB=132, P=8 (TS 38.214 Table 5.1.2.1.1-1: N_RB 73–144 → P=8): \(\lceil 132/8 \rceil = 17\) RBGs → 17-bit bitmap. Flexible non-contiguous allocation.
• Type 1: RIV (Resource Indication Value) — compact encoding of start RB + length for contiguous allocations. \(\text{RIV} = N_{RB}(L_{RBs}-1) + RB_{start}\) if \(L_{RBs}-1 \leq \lfloor N_{RB}/2 \rfloor\).

15.5 — Per-Slot Scheduling Pipeline

Full scheduling pipeline for our 128-UE DDDSUUDDD system (runs every DL slot = 0.5 ms):

15.6 — Peak Throughput Calculation

Single UE peak (SU-MIMO, rank-2, full BW, MCS 27):

From TS 38.214 Table 5.1.3.1-2 (256-QAM): MCS 27 → \(Q_m = 8\) bits/symbol, \(R = 948/1024\). Available symbols after overhead: 14 symb/slot − 2 DMRS − 1 PDCCH = 11 data symbols. Available subcarriers: 132 PRBs × 12 SC = 1,584 SC/symbol.

System sum throughput (MU-MIMO, 8 UEs × rank-2, full BW, MCS 27):

With spatial multiplexing of 8 UEs sharing the same 132 PRBs, the gNB delivers 4.1 Gbps aggregate in a single 100 MHz carrier — 8× over single-UE SU-MIMO. (Note: real deployments derate by ~15–25% for overhead, channel estimation error, and non-ideal precoding, giving ~3–3.5 Gbps practical peak.)

15.7 — HARQ Timing in DDDSUUDDD TDD Pattern

TDD pattern: D D D S U U D D D — slots 0,1,2,6,7,8 are DL; slot 3 is Special (DL+Guard+UL); slots 4,5 are UL. The HARQ-ACK feedback must arrive in a UL slot after the DL PDSCH. The timing indicator K1 in DCI specifies how many slots after PDSCH the UE transmits HARQ-ACK on PUCCH.

PDSCH in slots 0–2 benefits from quick HARQ-ACK (only 2 ms to UL slot 4), giving a short RTT of ~6 ms. PDSCH in DL slots 6–8 must wait for UL slot 13 (next 9-slot cycle), adding latency. With 128 UEs × 16 HARQ processes = 2,048 simultaneous HARQ contexts, the gNB must maintain the full HARQ state machine for all contexts at all times.

15.8 — PF Scheduler Snapshot (Plot)

Proportional Fair metric \(R_k(t)/\bar{R}_k(t)\) for 20 representative UEs in a single slot. Green bars: top-8 UEs selected for MU-MIMO scheduling. Grey bars: unselected UEs. The selection is driven by combined PF metric and spatial orthogonality.

15.9 — System Throughput vs MU-MIMO Layers (Plot)

Aggregate system throughput (Gbps) vs. number of simultaneously active spatial layers in MU-MIMO mode. Ideal linear scaling (blue dashed) assumes no IUI and perfect precoding. Practical curves (solid) account for: quantized PMI feedback loss, SRS estimation error, IUI residual with imperfect ZF, and PDCCH/DMRS overhead scaling. At 16 layers the gain vs ideal is ~60% due to feedback and interference effects.

15.10 — DCI Format 1_1 Field Structure

Each scheduled UE receives one DCI format 1_1 in the PDCCH, carrying all scheduling information for the PDSCH in that slot. Key fields:

15.11 — HARQ Combining and Redundancy Versions

5G NR uses HARQ-IR (Incremental Redundancy) via LDPC rate matching. The codeword is systematically punctured across four redundancy versions (RV0–RV3), each starting at a different point in the circular buffer. Combining at the UE:

Each redundancy version adds new parity information — IR combining provides approximately 1.5–3 dB per retransmission at low SNR (vs. chase combining 3 dB for identical retransmit). With 16 HARQ processes per UE, the gNB can have up to 16 in-flight PDSCH transmissions per UE simultaneously, maximizing pipeline utilization.

Functional Split Overview

3GPP TS 38.401 §6 defines eight functional split options for the NG-RAN architecture, ranging from Split 1 (between PDCP and RRC, leaving almost everything at the DU) to Split 8 (pure RF IQ sample transport — the full baseband stack sits centrally). The industry has converged on O-RAN Split 7-2x as the sweet-spot between centralisation gain and fronthaul burden.

Why 7-2x? Split 8 demands \(f_s \times 16\,\text{bit} \times 2\,(I{+}Q) \times 32\,\text{TRX} = 122.88 \times 10^6 \times 32 \times 32 \approx 125.8\,\text{Gbps}\) of raw fronthaul — infeasible over standard Ethernet. Split 7-2x moves the IFFT into the O-RU so only frequency-domain IQ (compressed) crosses the interface, and centralised baseband pooling becomes practical.

Fronthaul Bandwidth Calculation

Split 7-2x uncompressed (frequency domain IQ)

Per DL slot, the O-DU sends one frequency-domain IQ sample per subcarrier, per OFDM symbol, per antenna port:

With BFP-9 compression

Block Floating Point (BFP) groups 12 IQ samples per PRB under a shared 4-bit exponent; the mantissas are reduced to 9 bits effective. Compression ratio ≈ 16/9 ≈ 1.78×:

A 100GbE fronthaul link (after 64B/66B overhead ≈ 98.5 Gbps usable) comfortably accommodates this for a full 32-TRX configuration. In practice, many deployments use two bonded 25GbE links (totalling 50 Gbps raw) or a single 25GbE link with reduced antenna count (e.g., 8 TRX), since TDD asymmetry means DL and UL are not simultaneously at peak. For 8 TRX: \(45.4 / 4 \approx 11.4\,\text{Gbps}\) uncompressed, BFP-9 compressed ≈ 6.4 Gbps — fitting a 25GbE link with margin.

O-RAN CUS-Plane: Control, User, Synchronization

• C-plane (Control): O-DU → O-RU scheduling commands — which PRBs to use, beamforming weight indices, symbol timing, numerology. Must arrive at O-RU before the T1a_min deadline relative to the scheduled transmission time. C-plane travels via eCPRI message type 0 encapsulation.
• U-plane (User): frequency-domain IQ samples — DL: O-DU → O-RU (to be IFFT'd and transmitted); UL: O-RU → O-DU (FFT output from received PUSCH/PRACH). Compressed with BFP. Latency-critical: U-plane must arrive within the \([T_{2a,\min},\;T_{2a,\max}]\) window.
• S-plane (Sync): IEEE 1588v2 PTP (Precision Time Protocol) and/or SyncE (Synchronous Ethernet) distribute frequency and phase references. Target: \(\pm 130\,\text{ns}\) gNB-to-gNB phase alignment (O-RAN category B), enabling CoMP, TDD guard alignment, and accurate UL timing advance.

Fronthaul Latency Budget (O-RAN WG4 CUS §4)

The O-RAN WG4 CUS specification defines timing windows that bound when the O-DU must send data relative to the O-RU's scheduled transmission/reception time. All values below are for the category B low-latency profile.

The effective one-way transport latency budget follows from the U-plane window width: \[ \Delta t_\text{transport} < \frac{T_{2a,\max} - T_{2a,\min}}{2} = \frac{196 - 50}{2} \approx 73\,\mu\text{s} \] For a co-located O-DU/O-RU on the same rack this is trivial; for a remote O-RU at a cell tower it mandates a dark-fibre or dedicated Ethernet path with \(\leq 15\,\text{km}\) optical reach (round-trip light delay ≈ 100 μs per 10 km of fibre).

Block Floating Point (BFP) Compression

BFP exploits the fact that adjacent subcarriers in one PRB have similar magnitude (correlated channel energy). Per PRB (12 subcarriers):

1. Find the maximum magnitude across the 24 I/Q values in the PRB.
2. Express the common scale as a 4-bit exponent (shared header).
3. Store each sample as a reduced-width mantissa (9 or 10 bits).

Effective bit rate per sample with 9-bit mantissa + 4-bit exponent per 24 values:

O-RAN supports multiple compression modes: no compression, BFP (most common), μ-law, and beam-space compression (compresses in beam domain for massive-MIMO RUs). Compression type and exponent width are negotiated in the M-plane at O-RU startup.

eCPRI Protocol Basics

• Runs over Ethernet (IEEE 802.3); uses 64B/66B line coding.
• eCPRI header: 4 bytes — message type (8b), payload size (16b), PC_ID / sequence number.
• Message type 0: IQ data (U-plane), type 2: real-time control (C-plane).
• Minimum encapsulation overhead: Ethernet (18B) + eCPRI (4B) = 22B per frame; for typical 1-PRB U-plane payload ≈ 28 bytes, overhead fraction is notable — hence grouping multiple PRBs per eCPRI packet is standard practice (frequency-domain IQ grouped per symbol, per eAxC).
• eAxC (extended Antenna Carrier): identifies a specific (RU port, carrier, direction) stream; for 32 TRX → 32 DL eAxC IDs + 32 UL eAxC IDs = 64 streams.

Specifications: O-RAN.WG4.CUS §4 TS 38.401 §6 eCPRI spec v2.0

Phase 1 — Initial Cell Search

A UE (nodeB) powering on in a 5G NR cell knows nothing about the network. Cell search proceeds through a fixed sequence governed by the SSB (SS/PBCH Block):

1. GSCN scan: UE sweeps the Global Synchronization Channel Number grid (spaced 1.44 MHz on FR1) looking for the Primary Synchronization Signal (PSS). PSS is a length-127 m-sequence in the centre 132 subcarriers of the SSB. PSS correlation gives N_ID2 ∈ {0,1,2} and coarse time/frequency sync.
2. SSS detection: Secondary Synchronization Signal (Gold sequence) encodes N_ID1 ∈ {0..335}. \[N_{\text{cell\_ID}} = 3 \cdot N_{ID1} + N_{ID2} \in \{0\ldots 1007\}\]
3. PBCH decode: Physical Broadcast Channel delivers the MIB (Master Information Block): SFN[9:4], \(k_{SSB}\) (SSB subcarrier offset), CORESET#0 configuration, and \(\mu\) of common control.
4. SIB1 receive: gNB (nodeA) schedules SIB1 in CORESET#0 every 160 ms. UE decodes Type0-PDCCH to find SIB1 PDSCH → gets PLMN list, RACH configuration (preamble root, occasion times), TAC, cell barring status.

Phase 2 — RACH Procedure (4-message Handshake)

In DDDSUUDDD (slots 0–8 per 4.5 ms pattern), PRACH can only occupy the UL slots: slot 4 (first UL) and slot 5 (second UL), or UL symbols within the special slot (slot 3). With SCS 30 kHz, each slot = 0.5 ms, so slot 4 starts at \(t = 2.0\,\text{ms}\) in the pattern.

Msg1 — PRACH (slot 4, t = 2.0 ms)

The UE randomly selects a preamble index (0–63) from a root ZC sequence set configured in SIB1. The Zadoff-Chu root index \(u\) and cyclic shift \(N_{cs}\) determine the preamble set size. UE transmits PRACH format 0 (or short format C0/C2 for μ=1) in the configured PRACH occasion. logsA at nodeA:

logsA [slot= 4]: RACH_DET prn=37 ta=128 rsrp=-82dBm rnti_tmp=0xC3A1 slot=4

The detected timing advance ta=128 corresponds to a one-way propagation delay of: \[\tau = \frac{TA \times T_s}{2} = \frac{128}{2 \times 122.88 \times 10^6} \approx 0.52\,\mu\text{s} \quad(\approx 156\,\text{m})\]

Msg2 — RAR, Random Access Response (slot 6, t = 3.0 ms)

The gNB must transmit the RAR within the RA-Response window (configured 1–10 slots after Msg1). In DDDSUUDDD, the first DL slot after slot 4 is slot 6 (the next DL burst starts at slot 6 of the same pattern, t = 3.0 ms). The RAR PDSCH carries:

• TA command (11 bits): applied by UE to advance UL timing.
• UL grant: resource allocation for Msg3 PUSCH (frequency hopping may apply).
• Temporary C-RNTI: assigned for contention resolution.

logsA [slot= 6]: RAR_TX rnti=0xC3A1 ta_cmd=128 ul_grant=prb12..43 t_crnti=0xC3A1

Msg3 — RRC Setup Request (slot 13, t = 6.5 ms)

UE transmits Msg3 on PUSCH using the UL grant from RAR, after applying the TA command. Msg3 contains RRC Setup Request with UE identity (S-TMSI or random value for contention resolution).

Important TDD constraint: in DDDSUUDDD, slots 7 and 8 are Downlink slots — the UE cannot transmit PUSCH on them. The RAR is received in slot 6 (DL, t=3.0 ms); the next available UL slots are slot 13 and 14 of the next 9-slot period (t = 6.5 ms and 7.0 ms). The gNB UL grant in the RAR points to slot 13 (K2 = 7 relative to slot 6).

logsA [slot=13]: PUSCH_DEC rnti=0xC3A1 prb=12..43 mcs=4 rv=0 crc=PASS
logsA [slot=13]: MSG3_RX  ue_id=0x3F2A09C1 (random)

Msg4 — RRC Setup / Contention Resolution (slot 15, t = 7.5 ms)

gNB sends Msg4 on PDSCH addressed by the temporary C-RNTI. Msg4 echoes the UE identity from Msg3 for contention resolution — only the UE that sees its own identity transitions to RRC Connected; others reset and retry with random backoff. The earliest DL slot after gNB decodes Msg3 (slot 13) is slot 15 (= pattern slot 6 of the second period, t = 7.5 ms).

logsA [slot=15]: PDSCH_TX rnti=0xC3A1 msg4 ue_id=0x3F2A09C1 → CONTENTION_RESOLVED

Phase 3 — Data Flow: 128-UE Scenario

Downlink pipeline

1. IP packet arrives at gNB → SDAP header → PDCP (ciphering/integrity) → RLC (segmentation into SDUs) → MAC PDU assembly with BSR/PHR subheaders.
2. MAC scheduler evaluates all 128 connected UEs using Proportional Fair (PF) metric: \[M_k = \frac{R_k(t)}{\bar{R}_k(t)}\] where \(R_k(t)\) is the instantaneous achievable rate and \(\bar{R}_k(t)\) is the exponential moving average throughput. Up to 8 UEs are co-scheduled in MU-MIMO per slot.
3. DCI format 1_1 (for DL) generated per scheduled UE → mapped to CORESET → PDCCH DMRS + CCE aggregation → transmitted in first 1–3 symbols.
4. PDSCH on allocated RBs with beamforming weight \(\mathbf{W}\) from latest SRS: \[\mathbf{y} = \mathbf{W}\,\mathbf{x} + \mathbf{n}\] Up to 16 spatial layers (rank-16 DL) across 8 co-scheduled UEs × up to 2 layers each.
5. UE ACK/NACK on PUCCH format 0 in next UL slot (slot 4 or 5).

logsA [slot= 0]: PDCCH rnti=1001 dci=1_1 prb=0..131 mcs=22 layers=2 bw_idx=3
logsA [slot= 0]: PDSCH rnti=1001 tb=45056bytes rv=0 harq_id=2
logsA [slot= 4]: PUCCH rnti=1001 fmt=0 harq=ACK

Uplink pipeline

1. UE has data in buffer → sends SR (Scheduling Request) on PUCCH format 0 in PUCCH resource configured by RRC.
2. gNB receives SR → allocates PUSCH via DCI format 0_1 in next DL slot.
3. UE transmits PUSCH in UL slot using allocated PRBs, MCS, and number of layers (up to 4 UL layers with SRS-based precoding).
4. O-RU performs FFT, CP removal, frequency-domain IQ → sent via U-plane to O-DU.
5. O-DU: channel estimation (DMRS) → equalization → LDPC decode → CRC check → HARQ combine if needed.

logsA [slot= 4]: SRS   ue=42 bw=44 hop=0 rsrp=-75dBm rank_est=2
logsA [slot= 5]: PUSCH ue=42 prb=12..43 mcs=15 rv=0 crc=PASS layers=2
logsB [slot= 4]: PUCCH tx harq=ACK
logsB [slot= 5]: PUSCH tx prb=12..43 mcs=15

Beamforming Update Cycle in DDDSUUDDD

With a 9-slot TDD pattern (period = 4.5 ms), the BF weight update cycle is:

• Slots 4–5 (UL): UEs transmit SRS on configured resources. gNB estimates uplink channel matrix \(\mathbf{H}_{UL}\).
• UL↔DL reciprocity: With calibrated TRX chains, \(\mathbf{H}_{DL} \approx \mathbf{H}_{UL}^T\). gNB computes SVD or Zero-Forcing precoder \(\mathbf{W} = \mathbf{H}^\dagger\).
• Slots 6–8 + 0–2 (DL): Updated precoder applied to PDSCH. Beamforming tracks channel at 222 Hz update rate (1/4.5 ms).
• Doppler tolerance: At pedestrian speeds (3 km/h, 2.6 GHz), \(f_d \approx 7\,\text{Hz}\) — well below 222 Hz update rate. At vehicular (120 km/h), \(f_d \approx 290\,\text{Hz}\) — marginal; additional Doppler compensation needed.

Full logsA / logsB Trace Example (steady-state, MU-MIMO)

logsA [slot=  0]: PDCCH rnti=1001 dci=1_1 prb=0..131  mcs=22 layers=2 bw_idx=3
logsA [slot=  0]: PDCCH rnti=1002 dci=1_1 prb=0..63   mcs=18 layers=2 bw_idx=1
logsA [slot=  0]: PDSCH rnti=1001 tb=45056bytes rv=0 harq_id=2
logsA [slot=  0]: PDSCH rnti=1002 tb=20480bytes rv=0 harq_id=0
logsA [slot=  0]: MU_MIMO paired=[1001,1002,1003,1004] rank_total=8

logsA [slot=  3]: PDCCH rnti=1001 dci=0_1 prb=12..43  mcs=15  ← UL grant in special
logsA [slot=  4]: PUCCH rnti=1001 fmt=0  harq=ACK
logsA [slot=  4]: PUCCH rnti=1002 fmt=0  harq=NACK     ← retransmit triggered
logsA [slot=  4]: SRS   ue=42  bw=44 hop=0 rsrp=-75dBm rank_est=2
logsA [slot=  4]: PRACH prn=37 ta=128 rsrp=-82dBm      ← new UE attaching
logsA [slot=  5]: PUSCH rnti=1001 prb=12..43 mcs=15 rv=0 crc=PASS
logsA [slot=  5]: PUSCH rnti=1003 prb=44..75 mcs=12 rv=0 crc=PASS

logsA [slot=  6]: PDCCH rnti=1001 dci=1_1 prb=0..131  mcs=23 layers=2 ← updated BF
logsA [slot=  6]: PDCCH rnti=1002 dci=1_1 prb=0..63   mcs=18 layers=2 rv=1  ← NACK retx
logsA [slot=  6]: RAR_TX rnti=0xC3A1 ta_cmd=128 ul_grant=prb12..43

logsB [slot=  0]: PDCCH decoded DCI 1_1 prb=0..131
logsB [slot=  0]: PDSCH decode OK  crc=PASS layers=2
logsB [slot=  4]: PUCCH tx harq=ACK
logsB [slot=  5]: PUSCH tx prb=12..43 mcs=15
  

Full PHY Pipeline Latency Breakdown

RACH Configuration Parameters

The collision probability with \(N_{UE}=128\) UEs each choosing uniformly from 64 preambles is approximately:

In practice, access attempts are staggered by the UE stack (IDLE → RACH trigger is spread over random back-off windows) so the effective simultaneous load is far lower than 128. The gNB also supports 2-step RACH (Msg1 + Msg3 combined in msgA) standardised in 3GPP Rel-16 to reduce latency and collision impact for large UE populations.

Specifications: TS 38.213 §8 TS 38.321 §5.1 TS 38.300 §8 (RRC states) TS 38.211 §7.4 (PRACH)

After DMRS, PT-RS and other reference signals are placed, the remaining resource elements (REs) carry data. The number of net data REs directly bounds the Transport Block Size (TBS). This calculator lets you sweep DMRS configurations and instantly see the impact on available capacity.

Timing Reference Model

Every O-RAN fronthaul timing parameter is defined relative to a single anchor point: T_ref, the moment of Over-The-Air (OTA) symbol transmission. T_ref is the "heartbeat" of the system — it is the instant at which the O-RU begins radiating the first sample of a scheduled OFDM symbol. All timing windows express how far in advance (negative offset) or how far after (positive offset) a message must arrive relative to T_ref.

Both O-DU and O-RU are independently synchronized to a common absolute time reference via the S-plane (IEEE 1588v2 PTP). Because both ends share the same T_ref, the O-DU can compute exactly when to dispatch C-plane or U-plane messages so that they arrive at the O-RU within the required timing windows, accounting for the one-way fronthaul transit delay τ_FH.

DL Timing Parameters — T1a Family

Downlink fronthaul uses two separate delivery windows: one for the C-plane (scheduling commands and beamforming weight indices) and one for the U-plane (compressed IQ data for the IFFT). These windows do not overlap — C-plane must arrive significantly earlier to give the O-RU time to configure its processing pipeline before the IQ data arrives.

T1a_max_cp_dl — Maximum advance for DL C-plane

The C-plane message for a DL symbol must arrive at the O-RU no earlier than T1a_max_cp_dl before the scheduled OTA transmission. If the message arrives before this deadline, the O-RU may discard it as stale. The O-DU therefore has a window defined by [−T1a_max_cp_dl, −T1a_min_cp_dl] relative to T_ref to deliver each C-plane section.

• Category A (high latency / relaxed fronthaul): T1a_max_cp_dl = 500 μs
• Category B (low latency / tight fronthaul): T1a_max_cp_dl = 1250 μs

The wider Category B window (1250 μs = 2.5 T_slot) reflects that low-latency deployments have predictable, stable fronthaul and can afford to send C-plane much earlier, keeping the O-RU's internal scheduling queue full.

T1a_min_cp_dl — Minimum advance for DL C-plane

The O-RU requires a minimum processing time after receiving the C-plane message before it can execute the corresponding symbol — this includes parsing the section headers, loading beamforming weight indices, and configuring the digital front-end. C-plane arriving later than −T1a_min_cp_dl is too late and will be discarded.

• Category A: T1a_min_cp_dl = 250 μs
• Category B: T1a_min_cp_dl = 285 μs

T1a_max_up — Maximum advance for DL U-plane

U-plane IQ data must arrive at the O-RU inside its IFFT scheduling buffer, but not so early that the buffer overflows (the O-RU drops data arriving before −T1a_max_up). The U-plane window is substantially tighter than the C-plane window because IQ data volumes are much larger and the O-RU buffer is shallow.

• Category A: T1a_max_up = 200 μs
• Category B: T1a_max_up = 196 μs

T1a_min_up — Minimum advance for DL U-plane

After receiving U-plane IQ, the O-RU needs at minimum T1a_min_up to perform the IFFT, add the cyclic prefix, apply digital pre-distortion (DPD), and ramp the PA. IQ arriving within −T1a_min_up of T_ref cannot be processed in time and is dropped — a missed symbol.

• Category A: T1a_min_up = 50 μs
• Category B: T1a_min_up = 50 μs

tAdv — O-DU Timing Advance Compensation

In a distributed system, the O-DU and O-RU share the same T_ref clock (via PTP), but the O-DU's messages must traverse a physical fronthaul path with one-way latency τ_FH. Without compensation, the O-DU would need to subtract τ_FH mentally from every timing calculation. Instead, O-RAN introduces tAdv: the O-DU shifts its entire transmission schedule forward by tAdv so that messages arrive at the O-RU at the correct time relative to T_ref.

Derivation

Let t = 0 be T_ref (OTA symbol start). The O-RU's IFFT window opens at t = −T1a_max_up = −196 μs (Category B). For IQ data to arrive exactly at that moment:

With τ_FH = 50 μs (10 km fiber at 2×10⁵ km/s):

More generally, tAdv is the additional pre-advance the O-DU applies relative to a nominal reference offset, configured via OAM (M-plane NETCONF) or measured via PTP delay requests:

The O-DU adjusts its local slot counter to start all FH transmissions tAdv earlier. If τ_FH changes (fiber rerouting, temperature- induced delay drift), tAdv must be updated — a mismatch creates a systematic timing offset that manifests as U-plane arriving consistently early or late at the O-RU.

UL Timing Parameters — Ta3 and Ta4

Uplink fronthaul timing measures the delay between the end of a received UL OFDM symbol at the O-RU antenna and the delivery of the corresponding compressed IQ U-plane packet to the O-DU. Two parameters bound this delay.

Ta3 — O-RU UL Turnaround Time

After the last sample of a UL symbol is received at the O-RU ADC, the O-RU must: apply AGC/BFP compression, packetize into eCPRI U-plane frames, and begin transmission on the fronthaul. Ta3 is the time between end of UL symbol and start of O-RU U-plane transmission on the fronthaul interface.

• Category B: Ta3_min = 25 μs (O-RU starts sending almost immediately)
• Category B: Ta3_max = 1100 μs (O-RU may buffer up to 1.1 ms before forwarding)

Ta4 — O-DU UL Receive Deadline

Ta4 is the time from end of UL symbol to completion of O-DU U-plane reception. It adds the fronthaul transit delay to Ta3:

• Category B: Ta4_min = 0 μs (no lower bound — O-DU must be ready immediately)
• Category B: Ta4_max = 2200 μs (O-DU must have all UL IQ within 2.2 ms of UL symbol end)

In our DDDSUUDDD pattern (T_slot = 500 μs), UL slots are slots 4 and 5 (t = 2.0 ms and t = 2.5 ms from subframe start). The earliest DL retransmit opportunity after slot 5 is slot 6 (t = 3.0 ms) — only 500 μs after slot 5 ends. For the O-DU to decode PUSCH and generate a HARQ-NACK within that window, the practical Ta4 must be well below 300 μs. Ta4_max = 2200 μs applies only to non-real-time delivery paths (e.g., UL for higher-layer logging); HARQ requires Ta4 < 500 μs in the DDDSUUDDD TDD configuration.

Complete Timing Parameter Table

Worked Example — 30 kHz / DDDSUUDDD / 10 km Fiber

Assumptions

• Fronthaul fiber: 10 km single-mode fiber (siteA deployment)
• τ_FH = 10 km ÷ (2×10⁵ km/s) = 50 μs one-way
• Profile: Category B
• T_slot = 500 μs (μ=1, SCS=30 kHz), T_symbol ≈ 33.33 μs, 32 TRX
• T_ref = 0 = start of OTA transmission for DL slot 0, symbol 0

DL Transmission Timeline

The O-DU must send C-plane no later than t = −T1a_min_cp_dl = −285 μs, and no earlier than t = −T1a_max_cp_dl − τ_FH = −1250 − 50 = −1300 μs relative to T_ref (accounting for transit so that arrival at O-RU is within window). For U-plane: send at t = −T1a_max_up − τ_FH = −196 − 50 = −246 μs so arrival at O-RU is at t = −196 μs (exactly at IFFT window open).

S-Plane: PTP Synchronization

The S-plane carries IEEE 1588v2 PTP (Precision Time Protocol) with the ITU-T G.8275.1 telecom profile. Its sole purpose is to align the absolute time reference across the O-DU and all connected O-RUs so that T_ref means the same instant at every node.

Sync Hierarchy

1. Grandmaster clock: GNSS-disciplined (GPS/GLONASS/ BeiDou) oscillator, typically at a primary reference time clock (PRTC) at the network core. Accuracy: ±100 ns to UTC.
2. Boundary clock: aggregation switch or telecom router in the midhaul/backhaul network; forwards PTP with <100 ns additional holdover error.
3. O-DU: receives PTP; maintains a local phase-locked oscillator disciplined by PTP; exports timing to the fronthaul Ethernet switch.
4. O-RU: receives PTP via the fronthaul switch; phase-locks its internal SyncE and PTP slave clock. This is the direct source of T_ref.

Maximum Time Error Budget

3GPP TS 38.104 requires that TDD cells align their OTA slot boundaries to within ±1.5 μs inter-site (to prevent UL reception from one cell colliding with DL transmission from an adjacent cell in the guard period). O-RAN category B tightens this to ±130 ns at the O-RU to support CoMP, null-steering, and accurate timing advance estimation.

If T_error exceeds ±130 ns at the O-RU, the resulting inter-site timing misalignment degrades CoMP gain, increases inter-cell interference in TDD guard periods, and causes HARQ failures on UEs near cell edges (their PUSCH arrives at the wrong O-RU receive window). In logsA this appears as elevated TA_ERROR_EXCEED events and unexplained PUSCH CRC failure spikes on specific O-RU ports.

Specifications: O-RAN.WG4.CUS §4 O-RAN.WG4.CUS §7 ITU-T G.8275.1 TS 38.104 §6.5 IEEE 1588v2

See §24 — O-RAN FH Complete Event Sequence for the full slot-by-slot timeline integrating C-plane scheduling, U-plane delivery, UL reception, HARQ decode, and retransmit windows across all DDDSUUDDD slot types.

For channel-by-channel CUS-plane section type mapping, see §25 O-RAN Channel Segregation.

Scenario: DL Slot 0 — 8 MU-MIMO UEs, Full Slot PDCCH + PDSCH

Complete Trace Log — logsA (O-DU/O-RU) and logsB (UE)

logsA t=-3500μs: SCHED_START slot=0 ues=8 mu_mimo=on layers=16 prb=132
logsA t=-3200μs: PRECODER_DONE W=32x16 svd_iter=14 spatial_compat_check=PASS
logsA t=-1250μs: CPLANE_TX slot=0 section_type=1 nsymbs=14 nprb=132 nports=32
logsA t=-1200μs: [O-RU RECV] CPLANE slot=0 stored
logsA t= -246μs: UPLANE_TX slot=0 sym=0 nprb=132 iq_size=45056B (PDCCH IQ)
logsA t= -196μs: [O-RU RECV] UPLANE slot=0 sym=0 ready
logsA t= -175μs: UPLANE_TX slot=0 sym=2 nprb=132 iq_size=101376B (PDSCH sym2 IQ)
logsA t=    0μs: OTA_TX slot=0 sym=0 (symbol 0 begins, T_ref)
logsA t=   37μs: OTA_TX slot=0 sym=1
logsA t=   73μs: OTA_TX slot=0 sym=2 (PDSCH begins)
logsA t=  500μs: OTA_TX slot=0 DONE (all 14 symbols complete)
logsA t= 1250μs: CPLANE_TX slot=5 section_type=1 nsymbs=14 nprb=66 nports=32 (UL slot 5)
logsA t= 1750μs: CPLANE_TX slot=6 section_type=1 nsymbs=14 nprb=132 nports=32 (HARQ retx speculative)
logsA t= 2000μs: OTA_RX slot=4 UL sym=0 (slot 4 UL starts at O-RU)
logsA t= 2033μs: UPLANE_RX slot=4 sym=12 nprb=6 (PUCCH IQ arriving at O-DU)
logsA t= 2083μs: PUCCH_RX rnti=0xABCD harq_ack=1 (ACK received at O-DU)
logsA t= 2150μs: HARQ_DEC pid=3 rnti=0xABCD result=ACK buf=free
logsA t= 2500μs: OTA_RX slot=5 UL sym=0 (slot 5 PUSCH at O-RU)
logsA t= 2550μs: UPLANE_RX slot=5 sym=0..13 nprb=66 (PUSCH IQ at O-DU)
logsA t= 2700μs: PUSCH_DECODE slot=5 rnti=1042 prb=0..65 mcs=16 rv=0 crc=PASS

logsB t=    0μs: PDCCH_DET slot=0 sym=0 (UE begins CORESET capture)
logsB t=   73μs: PDCCH_DECODE slot=0 dci_fmt=1_1 rnti=0xABCD mcs=22 prb=0..131 harqid=3 k1=4
logsB t=   73μs: PDSCH_RX_START slot=0 sym=2 (UE begins PDSCH symbol capture)
logsB t=  500μs: PDSCH_RX_DONE slot=0 (all 12 PDSCH symbols captured)
logsB t=  620μs: PDSCH_DECODE slot=0 rnti=0xABCD crc=PASS tb=45056bytes layers=2
logsB t= 2000μs: PUCCH_TX slot=4 sym=12 harq_ack=1 (ACK transmitted by UE)
logsB t= 3000μs: PDCCH_DET slot=6 (if NACK: retx DCI received, else idle)
  

Event Timeline — Full Sequence (t = −1300 μs to +3500 μs)

UL Timing Detail — PUCCH → HARQ Decision → Retransmit Window (t = +2000 to +3200 μs)

§25.1 — CUS-Plane Section Types Overview

The O-RAN fronthaul CUS-plane defines discrete Section Types that allow the O-DU to communicate per-symbol, per-PRB scheduling intent to the O-RU. Each physical channel in 5G NR is mapped to one specific Section Type based on its timing, bandwidth, and precoding characteristics. Understanding which Section Type carries which channel is the first step toward diagnosing fronthaul issues in logsA / logsB traces.

§25.2 — Per-Channel CUS-Plane Mapping Table

The table below maps every 5G NR physical channel to its O-RAN Section Type, the eAxC streams used, relevant symbol range, and operational notes specific to the μ=1 (30 kHz SCS), 100 MHz BW, 32TRX, 128-UE system reference.

§25.3 — eAxC ID Structure

An eAxC (extended Antenna-Carrier) ID is the O-RAN identifier that tags every fronthaul packet with its originating or destination antenna port and carrier. For a 32TRX single-cell system, the eAxC ID bitfield is constructed as follows:

§25.4 — DL vs UL Fronthaul Streams (C/U Separation)

§25.5 — PRACH Special Handling (Type 3 Deep Dive)

PRACH occupies a unique position in the CUS-plane architecture. Because the preamble sequence (Zadoff-Chu root index, format, cyclic shift) is determined by the O-DU, but the reception window is a precise time-frequency resource within the uplink frame, the O-DU must send a dedicated Section Type 3 C-plane message to the O-RU ahead of every PRACH occasion.

[slot=4 sym=0] PRACH_IQ
  section_type=3
  eAxC=0x0040   (PRACH port 0)
  timeOffset=0
  freqOffset=-528  (kHz from CC centre)
  nRb=6
  format=0       (LRA=839)
  startPrbc=63   (centre of 132 PRB grid)
  guardband=2
  compMethod=NONE (PRACH uncompressed)

[slot=4 sym=0..13] PUSCH_IQ
  section_type=1
  eAxC=0x0000..0x001F  (ports 0-31)
  startPrbc=0 numPrbc=132
  numSymbol=14
  compMethod=BFP udIqWidth=9
      

FH Bandwidth Distribution chart: