Engineering Theory: O-RAN Split 7-2x & eCPRI Fronthaul Sizing
1. The Evolution of Mobile Fronthaul: From Legacy CPRI to eCPRI
In traditional 4G LTE Centralized RAN (C-RAN) deployments, the Common Public Radio Interface (CPRI) utilized the Option 8 functional split, in which digitized time-domain I/Q samples were transmitted directly between the Baseband Unit (BBU) and the Remote Radio Head (RRH) after analog-to-digital conversion (ADC/DAC). Because Option 8 digitizes the full analog bandwidth regardless of cell traffic or subcarrier occupancy, the required CPRI bit rate was constant, continuous, and strictly governed by:
BWCPRI = 2 × Fs × Nbits × Nantennas × (16/15)Control × (66/64)LineCodingWhere
Fs is the baseband sampling rate (e.g., 30.72 MSPS for 20 MHz LTE), Nbits is sample resolution (15 or 16 bits per component), and Nantennas is the physical antenna port count.
While manageable for 2T2R and 4T4R 20 MHz LTE (requiring ~2.5 to ~9.8 Gbps), applying Option 8 to a 5G NR 100 MHz channel with 64T64R Massive MIMO results in a transport demand exceeding 140 Gbps to 240 Gbps per sector. Transporting ~450 Gbps to ~720 Gbps across a 3-sector macro site would require dozens of dedicated 25G/100G fiber strands per tower, making centralized cloud architectures commercially and physically unviable.
To overcome this physical bottleneck, the eCPRI Consortium and the O-RAN Alliance Working Group 4 (O-RAN WG4) standardized the Split Option 7-2x architecture. By moving the lower Physical layer functions—specifically the Fast Fourier Transform (FFT/iFFT), Cyclic Prefix (CP) insertion/removal, and digital beamforming—into the Open Radio Unit (O-RU), the transport interface moves from the time domain to the frequency domain.
2. The O-RAN Split 7-2x Architecture (O-RAN.WG4.CUS.0)
Under the O-RAN Split 7-2x specification, baseband processing is split cleanly across two primary network entities:
- O-DU (Open Distributed Unit): Handles the upper physical layer, including channel coding (LDPC for data, Polar codes for control), rate matching, modulation mapping (QPSK up to 256-QAM), layer mapping, channel estimation, equalization, and scrambling.
- O-RU (Open Radio Unit): Executes digital beamforming weighting, subcarrier de-mapping, resource element (RE) extraction, 4096-point FFT/iFFT, cyclic prefix insertion, analog radio frequency (RF) conversion, power amplification, and filtering.
Category A vs. Category B O-RU Implementations: In Category A O-RUs, digital beamforming occurs in the O-DU, requiring separate I/Q streams per antenna polarization. In Category B O-RUs (Standard Massive MIMO), digital beamforming weights are applied directly within the O-RU. Consequently, the fronthaul bandwidth is governed not by the 64 physical transceivers, but by the number of simultaneous spatial transmission layers (typically up to 16 downlink layers and 8 uplink layers in multi-user MIMO).
Crucially, Split 7-2x exhibits traffic-dependent bandwidth scaling. If a cell site is idling during late-night hours with only 10% of Physical Resource Blocks (PRBs) scheduled, eCPRI user-plane transmission drops by 90%, enabling tremendous statistical multiplexing gains across shared optical rings and metro edge packet switches.
3. Block Floating Point (BFP) IQ Compression Algorithms
Even with Split 7-2x moving the boundary to the frequency domain, uncompressed 16-bit linear I/Q samples (32 bits per complex sample) for 100 MHz 16-layer Massive MIMO still require approximately 73.3 Gbps per O-RU, immediately saturating standard 25GE optical transceivers. To resolve this, O-RAN WG4 defines standardized Block Floating Point (BFP) compression:
- Mathematical Principle: BFP compression groups the 12 complex Resource Elements within each Resource Block (PRB). The algorithm evaluates all 12 complex samples, identifies the maximum absolute amplitude among them, and computes a single 4-bit shared exponent for the entire PRB.
- Mantissa Quantization: Each individual sample's in-phase (I) and quadrature (Q) components are then scaled and quantized to a compact 9-bit mantissa (BFP-9) or 14-bit mantissa (BFP-14).
-
Spectral Integrity & EVM: BFP-9 reduces the effective payload from 32 bits to just
9 + 9 + (4/12) = 18.33 bitsper Resource Element, representing an immediate 43.7% reduction in raw transport bit rate. Because the dynamic range is preserved by the shared exponent, the Error Vector Magnitude (EVM) degradation is lower than 0.5%, comfortably satisfying the 3GPP TS 38.104 256-QAM RF transmission compliance limit of 3.5% EVM.
4. Optical Port Sizing, Dark Fiber Leasing & Techno-Commercial Economics
In carrier-grade transport design, optical interfaces must be sized for the peak transmission burst rate during full downlink allocation to prevent internal queue buffer exhaustion, packet drop, and severe TCP throughput collapse.
Network planners must evaluate the optical interface threshold:
- The 25GE SFP28 85% Utilization Rule: While 25GE links support a theoretical line rate of 25.0 Gbps, Ethernet framing, inter-packet gap (IPG), forward error correction (FEC IEEE 802.3by RS-FEC), and burst buffer dynamics require an engineering ceiling of 85% sustained utilization (21.25 Gbps).
- Sizing Decision Matrix: When a 64T64R 100 MHz O-RU operates with BFP-9, its peak throughput reaches 41.25 Gbps. This requires two 25GE SFP28 links per O-RU (configured via link aggregation or eCPRI flow-splitting), or a single 100GE QSFP28 link.
- Dark Fiber Leasing Optimization: For a typical 3-sector macro site, 6 × 25GE links would require 12 dedicated fiber strands under standard duplex optics. By deploying Bi-Directional (BiDi) SFP28 transceivers operating at 1270nm/1330nm on a single strand, or WDM (CWDM/DWDM) passive muxponders, operators cut fiber lease expenses by 50% to 83%, saving thousands of dollars per tower site annually in municipal right-of-way and dark fiber IRU fees.