Deterministic Wireless Baseband: Full Technical Detail
These specifications describe the target architecture and design intent for Air Interface. Numbers are architecture estimates and design targets; final figures depend on implementation and validation, and capabilities described as design goals are not yet demonstrated on hardware.
| General | |
| Product Class | Common digital baseband SoC for fixed wireless, backhaul, tactical, and industrial radio |
| Frequency Range | 900 MHz to 80+ GHz (band-specific RF personality module) |
| Channel Bandwidth | Sub-7 GHz: 1.5 to 160 MHz per carrier, up to 4 carriers for 640 MHz aggregated (NC-CA) 24 GHz: 100 to 800 MHz V-band / E-band: 250 to 2160 MHz (RF-module dependent) |
| Duplex | TDD (default) + FDD (paired-spectrum option, e.g. 11 GHz, 18 GHz, E-band) |
| Topology | PTMP (sector/cell), PTP (backhaul), Mesh (multi-hop with elected coordinator) |
| Deliverable | Baseband SoC + reference design package (schematics, driver, daemon, CLI, test suite, RF reference designs) |
| PHY | |
| Waveform | CP-OFDM / OFDMA (DL and UL). Optional DFT-s-OFDM for UL on power-limited devices. |
| Numerology Family | Indexed by parameter mu (0 to 6). SCS = 15 kHz x 2^mu. Range: 15 kHz to 960 kHz. |
| FFT Sizes | Multiple sizes supported, selected per profile to fit target bandwidth and SCS. |
| Modulation | QPSK through 4096-QAM (profile-gated). Ultra-QAM requires Snapshot DPD support (ORX lane + coefficient updates) and pilot-anchored coherence health. |
| Peak Modulation Order | 12 bits/symbol (4096-QAM). Net bps/Hz depends on code rate and overhead (pilots/guards) and is profile-reported. |
| Carrier Aggregation | Non-contiguous carrier aggregation (NC-CA), up to 4 independent fragments. The scheduler supports per-carrier timelines and per-carrier regulatory pauses. |
| Virtual Radio Slices | 4 independent PHY slices (4x4 units) that can be ganged or split. Elastic chain mapping trades spatial depth for frequency width: 16x16 @ 160 MHz, 8x8 @ 320 MHz, or 4x4 @ 640 MHz aggregated. |
| FEC | QC-LDPC with rate matching. Code rates from 1/3 through 8/9. |
| Control Channel FEC | Robust short-block coding with CRC (profile selectable) |
| HARQ | Multi-process with soft combining. Deterministic ACK/NACK scheduling. |
| MIMO / Beamforming | |
| RF Chains | 4x4 to 16x16 per chip. Multi-chip clustering for higher-order arrays. |
| Modes | SU-MIMO (multi-layer to one user), MU-MIMO (spatial separation, multiple users), Hybrid beamforming (mmWave analog + digital) |
| CSI Acquisition | TDD reciprocity (preferred) or explicit feedback |
| Precoding | Quality-gated: automatically reverts to safe defaults if calibration is insufficient. |
| Endpoint Model | Meaningful antenna resources at both base and remote nodes. Both ends contribute to spatial separation and interference suppression. |
| MAC / Scheduler | |
| Scheduling | Hardware Scheduler Engine (HSE) in silicon. MAP-driven, deterministic. |
| Frame Duration | 1 ms nominal. Mini-slots supported for ultra-low latency. |
| Resource Grid | OFDMA: PRB allocated in RU quanta per profile. |
| Contention | Scheduled by default. Optional contention windows for initial access and mesh discovery. |
| Interference | |
| Resilience | Adaptive interference mitigation, link-quality validation, and graceful recovery behavior. |
| Recovery Mode | Protected low-rate operation preserves coordination when normal throughput cannot be sustained. |
| Telemetry | Operator-visible link-health indicators reflect real deployment conditions. |
| Physical / Electrical | |
| Operating Temperature | -40 C to +85 C (ambient) |
| Power (Baseband) | Pre-production estimate. Final specifications at product launch. |
| Host Interface | PCIe or AXI. DMA engines with timestamped interrupt delivery. |
| RF Interface | Variant-dependent: JESD204C Subclass 1 (virtual ports, dynamic lane mapping) or integrated data converters. Optional ORX observation lane (RF-module dependent) for Snapshot DPD. |
| Management Plane | gRPC, REST, CLI via host-side daemon. Outside real-time path. |
| Timing | 1PPS/PTP hard-lock for UTC-aligned frame boundaries. Holdover and OTA sync options preserve TDD alignment during outages. |
| Security | Hardware-accelerated encryption. Hitless re-key. Hardware root of trust. |
CP-OFDM / OFDMA across all bands
Air Interface parameterizes one waveform family, CP-OFDM with OFDMA, across every operating band. Subcarrier spacing, FFT size, cyclic prefix length, and pilot density are profile-selectable. A 900 MHz part and a 60 GHz part differ in profile parameters and front-end bandwidth, not in architecture: the processing pipeline is the same across the family.
| DL Waveform | CP-OFDM with OFDMA resource allocation |
| UL Waveform | CP-OFDM (default) or DFT-s-OFDM (optional, power-limited devices) |
| Resource Grid | Time-frequency grid. PRB = 12 subcarriers x 1 OFDM symbol. RU = N_PRB x N_SYM x N_LAYER. |
| Occupied Spectrum | ~85% of FFT bins carry data or pilots; remainder reserved for DC, guard bands (occupancy factor ~0.85). Net payload throughput is further reduced by pilot overhead, CP, control/MAP, HARQ, and calibration reservations. |
| PAPR Mitigation | DFT-s-OFDM option for UL. Optional clipping/windowing for DL. |
Subcarrier spacing, symbol timing, cyclic prefix
Subcarrier spacing scales exponentially (mu = 0 to 6, 15 kHz to 960 kHz), assigned per band profile: longer symbols for severe multipath, shorter symbols against phase noise and high Doppler.
Validated reference configurations across the spectrum
Air Interface defines reference profiles for Sub-GHz (900 MHz), Mid-Band (CBRS / 3.5 GHz), UNII (5 to 7.1 GHz), and mmWave (24 GHz, V-band, E-band) deployments. Integrators can use these reference configurations or parameterize custom profiles within the supported 15 kHz to 960 kHz subcarrier spacing bounds.
Representative PHY-layer capacity per profile
The tables below show representative PHY throughput under sustained conditions adequate for the chosen modulation. Representative rows use conservative 1-2 layer service profiles; peak silicon stream count is higher and profile-dependent. Sub-7 GHz profiles can enable Ultra-QAM when the RF module supports Snapshot DPD (observation receiver + periodic coefficient updates). mmWave profiles assume 2x2 dual-polarization MIMO, which is standard for line-of-sight backhaul and mesh.
Air Interface keeps silicon cost and power bounded by trading spatial depth for frequency width. The chip exposes 16 logical streams that can be ganged on one carrier or split across multiple non-contiguous carriers.
| Mode | Carriers | BW / Carrier | Total BW | Digital Streams / Carrier | Use Case |
|---|---|---|---|---|---|
| A. The Shield | 1 | 160 MHz | 160 MHz | 16 | Urban/noisy: maximum interference rejection (deep nulls) |
| B. The Backhaul | 2 | 160 MHz | 320 MHz | 8 | Balanced: high capacity with strong diversity |
| C. The Scavenger | 4 | 160 MHz | 640 MHz | 4 | Rural/clean: maximum throughput using fragmented spectrum |
These are PHY-layer planning estimates. Net payload throughput is lower after pilots, cyclic prefix, control, scheduler overhead, and calibration windows. Ultra-QAM (1024/4096-QAM) requires Snapshot DPD support in the RF module and sufficient SINR; if DPD or coherence health is degraded, the system fails closed to a safer MCS. mmWave figures assume 2x2 dual-polarization MIMO, standard for LoS backhaul at these frequencies.
The baseband supports QPSK through 4096-QAM (profile-gated). For Ultra-QAM profiles, proprietary Snapshot DPD linearizes the transmit chain when the attached RF module provides an observation path (ORX). Rate control promotes to higher modulation only when link quality confirms sufficient margin. MCS caps per band profile allow operators to limit maximum modulation order if desired.
| QPSK | 2 bits/symbol. Robust fallback for low-SINR, long-range, or hostile environments. |
| 16-QAM | 4 bits/symbol. Coverage-layer workhorse. |
| 64-QAM | 6 bits/symbol. Standard operating point for moderate SINR fixed links. Conservative wideband ceiling for the longest mmWave links; narrower, cleaner mmWave channels can sustain 256-QAM. |
| 256-QAM | 8 bits/symbol. High-capacity point for good fixed links. Achievable at mmWave with narrow channels and high SNR. |
| 1024-QAM | 10 bits/symbol. Snapshot-DPD-assisted. Short to medium range sub-7 GHz links with calibrated RF. |
| 4096-QAM | 12 bits/symbol. Snapshot-DPD-assisted. Ultra-QAM for high-SINR calibrated links; net bps/Hz depends on code rate and overhead. |
To make 1024-QAM and 4096-QAM realistic on an infrastructure power budget, Air Interface uses proprietary snapshot-based digital pre-distortion.
| Algorithm | Proprietary snapshot-based digital pre-distortion |
| Observation Path | 1x ORX lane per RF sector (RF module dependent) |
| Targets | Ultra-QAM-class linearity for 1024-QAM and 4096-QAM profiles (when calibrated) |
| Fail-Closed | If ORX/DPD health is degraded or coefficients are stale, Ultra-QAM is gated off and the scheduler falls back to safer MCS. |
Primary FEC is quasi-cyclic LDPC with rate matching, chosen for hardware parallelism, mature decoder implementations, and strong performance across block sizes. The system supports code rates from 1/3 (maximum protection) through 8/9 (maximum throughput), with incremental redundancy for HARQ soft combining.
Control channels use a separate, more robust coding scheme optimized for reliable decode under worst-case SINR before data can be scheduled.
| FEC Type | QC-LDPC (data channels), robust short-block coding (control channels) |
| Code Rates | 1/3, 1/2, 2/3, 3/4, 5/6, 8/9 |
| HARQ | Multi-process with incremental redundancy and soft combining. Deterministic ACK/NACK scheduling. |
| Max Retransmissions | Profile-configurable |
Air Interface family members support 4x4 to 16x16 coherent RF chains, sized to the target deployment. For applications requiring higher-order arrays, multiple chips can be clustered with synchronized timing. Both ends of the link carry meaningful antenna resources, enabling spatial separation and interference suppression in both directions. mmWave profiles use 2x2 dual-polarization as the standard configuration for line-of-sight backhaul.
| Single Chip | 4x4 to 16x16 coherent RF chains. Configuration selected at integration time. |
| Clustering | Multi-chip synchronization for higher-order arrays (32x32 and beyond). Shared timebase and coordinated scheduling. |
| mmWave Standard | 2x2 dual-polarization (XPIC). Standard for LoS backhaul and mesh at V-band and E-band. |
| SU-MIMO | Multiple spatial layers to a single user. |
| MU-MIMO | Multiple users share time/frequency resources via spatial separation. Candidate groups prepared at slow timescale; hardware selects per allocation at fast timescale. |
| Hybrid Beamforming | mmWave: analog beam steering plus digital precoding for residual spatial separation. |
| CSI: TDD Reciprocity | Preferred for fixed wireless and mmWave. UL sounding provides DL channel estimate. Requires per-chain RF calibration. |
| CSI: Explicit Feedback | Quantized channel feedback when reciprocity is unavailable. Periodicity managed by scheduler. |
| Precoding Quality Gate | MU precoding is quality-gated. If calibration or channel confidence is insufficient, the system reverts to safe single-user defaults. Fail-closed by design. |
| Sub-Band Processing | Beamforming and combining are computed per sub-band for frequency-selective spatial reuse. |
| Beam Management | mmWave: beam sweep during initial access, continuous beam tracking in steady state. Training overhead bounded and accounted for by scheduler. |
Resilient operation in congested and contested spectrum.
Air Interface combines adaptive interference mitigation, link-quality validation, and graceful recovery behavior to keep networks operating in congested and contested RF conditions.
The system is built for real deployment problems: same-channel neighbors, sector bleed, bursty overlap, rooftop backhaul spillover, and irregular coexistence events in shared spectrum. Rather than relying on a single lab-only rejection number, Air Interface reports practical operating margin under current link conditions and adapts service accordingly.
When normal throughput cannot be sustained, the link transitions to a protected recovery mode designed to preserve coordination and shorten time-to-recovery.
Deterministic scheduling in silicon
| Timing | |
| Frame Duration | 1 ms nominal |
| Mini-Slot Support | Sub-frame scheduling for ultra-low latency applications |
| TDD Pattern | Per-frame DL/UL split, scheduler-decided. Guard/ramp overhead at TX/RX transitions. |
| Resource Allocation | |
| Allocation Quantum | OFDMA resource units, sized per profile to bound hardware scheduling runtime. |
| MAP Generation | Hardware generates per-frame MAP describing all allocations. |
| Multi-Carrier (NC-CA) | Up to 4 parallel per-carrier timelines (multi-vector scheduling). Per-carrier DFS/LBT pauses do not stall other carriers; MAC sees one aggregated service. |
| MU Scheduling | Two-stage: candidate groups prepared at slow timescale, hardware selects per allocation at fast timescale. |
| Capacity | |
| Data Capacity | Available capacity per frame accounts for control overhead and calibration reservations. Both are bounded and reported in utilization counters. |
| Calibration | Dynamic windows for PHY sounding and calibration. Scheduler-managed, transparent to the host. |
| Mesh Support | |
| Coordinator | Elected or rotated, with deterministic fallback when coordination is degraded. |
| Multi-Hop | Scheduled relay with per-hop allocations. No random contention in the forwarding path. |
| Hardware Interface | |
| Bus | PCIe or AXI (integration dependent). For >10 Gbps profiles, PCIe Gen4 x4 (or Gen3 x8) is recommended. |
| Configuration | All operating parameters (duplex mode, band profile, scheduling policy, interference response) configurable from the host. Band changes require no firmware rebuild. |
| Data Movement | DMA engines with timestamped interrupt delivery |
| Software | |
| OS Integration | Linux kernel driver (source provided). Standard netdev interface, standard sockets, standard routing daemons. |
| Management Daemon | User-space daemon exposing gRPC, REST, and CLI. Device configuration, profile selection, telemetry export. |
| API Documentation | Included in reference design package (NDA required) |
| Observability | |
| Telemetry | Structured export from every subsystem: scheduler utilization, per-link signal quality, interference environment, thermal state, error counters, fault escalation. |
| Diagnostics | Live schedule and PHY state capture for field troubleshooting without interrupting the dataplane. |
| Manufacturing | Built-in test and loopback modes for production bring-up. Factory unlock required. |
| Internal Timebase | 64-bit nanosecond-resolution counter (hardware) |
| 1PPS Hard-Lock | HSE phase-locks the FRAME_START event directly to GNSS 1PPS or PTP sync pulse edge (phase alignment, not just frequency discipline). |
| Holdover | On loss of 1PPS, local oscillator maintains frame boundary precision for extended periods (profile/board dependent). |
| Disciplining Sources | Primary: GNSS 1PPS. Secondary: IEEE-1588v2 PTP (slave). Tertiary: OTA sync (remotes recover timing from base preamble). |
| Frame Events | Hardware-generated events at frame, slot, and symbol boundaries (deterministic) |
| TDD Switching | All TX/RX transitions referenced to the timebase. Guard times per transition are profile-defined. |
| Network-Wide Phase | All nodes share the same UTC-aligned frame boundary, eliminating co-location hidden-node interference between adjacent sectors. |
| Operating Temperature | -40 C to +85 C (ambient) |
| Power (Baseband) | Pre-production estimate. Final specifications at product launch. |
| Thermal | Lidless FC-BGA. Chassis-bond recommended for outdoor deployments. |
| RF Interface | Variant-dependent: JESD204C Subclass 1 (virtual ports, dynamic lane mapping) or integrated data converters. Optional ORX observation lane (RF-module dependent) for Snapshot DPD. |
| Split Architecture | Digital baseband can be placed independently of RF personality module (indoor/outdoor split, rack-mount with remote head, etc.). Interface specification available under NDA. |
| Security | Hardware-accelerated encryption. Hitless re-key. Fail-closed architecture. Hardware root of trust. |
