Technical evaluation · for RAN / PHY engineers
Tamper-evident channel state, for the cell edge.
For RAN PHY engineers at cellular operators evaluating performance, compliance, and forensic posture. We extract the channel state M t in real time, seal it into a transition hash chain bound to an FPGA-resident PUF key, and use the verified prior to improve cell-edge BER on the 3GPP conformance channels. All measurements below were produced by the reproducible Python pipeline linked at the bottom of this page — no proprietary data, no cherry-picked SNRs.
Request engineering briefing Read the whitepaper
§0 · Reader's note — what is and is not claimed
Two independent value props on this page. Evaluate them separately.
(1) Cell-edge BER reduction (§3 – §6). Channel-aware reception with known/estimated CSI on 3GPP conformance channels (EPA / EVA / Pedestrian-A / Vehicular-A) is well-trodden territory — MMSE-LE / DFE / OFDM equalizers have been the production baseline for two decades. The licensable contribution here is not the equalizer; it is the verified-prior loop in §6 that lets the equalizer's M t anchor against a chain-trusted past. The 40% worst-step BER reduction shown is over cell-edge handover with a synthetic CSI degradation profile and a SISO chain; pilot integration with OFDM / MIMO PHY is where real throughput numbers come from.
(2) Tamper-evident channel-state log (§7). The FPGA-PUF sealed transition hash chain over a cellular channel-state log is the standalone licensable asset. Lawful-intercept compliance (KCC / FCC / ETSI), enterprise SLA forensics, and inter-operator roaming settlement all depend on a tamper-evident shared ground truth of network behaviour — which the chain provides without trusting either operator. This claim applies regardless of which equalizer or PHY produced M t .
A PHY engineer who is sceptical of (1) on well-known-equalization grounds can still adopt (2) standalone — the chain primitive is PHY-agnostic.
⚡ Signal / Data Restoration
Broken CSI under cell-edge handover → channel state reconstructed.
Patent core claim: "recover original modal information from broken received signal via medium forward model + integrity-anchored prior" . Cell-edge fading and roaming hand-offs collapse standard channel estimation. The M t forward model + chain-verified prior reconstructs CSI across the handover boundary — 40% worst-step BER reduction (12.7% → 7.6%).
§1 · Why a cellular operator should read this
Two value axes: cell-edge throughput, and tamper-evident channel logs.
The receiver-side improvement (lower BER under the same E b /N 0 , on the same 3GPP conformance channels) is the well-known equalization gain — useful at the cell edge, where UEs spend a disproportionate fraction of their data minutes. The novel piece is the integrity seal: every channel-state snapshot M t is cryptographically chained to its predecessor and signed by a PUF key inside an FPGA. The signed chain is a regulator-grade evidentiary record of network behavior.
Cell-edge BLER
Reduce BLER at UE near cell edge — fewer HARQ retransmissions, higher MCS at same SNR.
Lawful intercept logs
Tamper-evident channel-state record for LI compliance (KCC, FCC, ETSI EU mandates).
SLA forensics
Independently verifiable evidence in enterprise SLA disputes — "was the drop our fault?".
Roaming settlement
Both operators can sign and verify the same M t chain — neutral evidence base.
§2 · Experimental setup
SISO QPSK, frequency-flat AWGN + frequency-selective multipath.
This page intentionally evaluates a SISO baseband chain — no MIMO, no OFDM, no HARQ. The point is to isolate channel-state-aware reception (Subsystem B of the underlying patent) from confounding gains. Production integration into an OFDM/MIMO PHY is a separate program of work; the same M t chain attaches to any PHY.
Modulation
QPSK, Gray-coded, unit symbol energy
Channel models
ITU-R Rec. M.1225 Vehicular-A / Pedestrian-A; 3GPP TS 36.101 Annex B EPA / EVA
Sample rate
10 Msps (100 ns sample period)
Noise
Complex AWGN, identical realization for both receivers within each trial
Sample size
200,000 bits per E b /N 0 per channel
Receiver A — no M t
Coherent QPSK with single-tap carrier recovery (no equalization)
Receiver B — with M t
Frequency-domain MMSE-LE, full channel response (perfect CSI assumed — see §8)
Theoretical bound
QPSK AWGN: P b = Q(√(2 E b /N 0 ))
§3 · Results · 3GPP / ITU-R conformance channels
Same equalization gain you already model in your link-level simulator.
Nothing surprising in this section — these are the gains a textbook MMSE-LE produces with perfect CSI on the channels you already validate against. The point of reproducing them is to ground the rest of the page in numbers any link-level engineer can independently verify.
Channel
Receiver A — BER
Receiver B — BER
Δ E b /N 0 to BER 10⁻³
ITU-R Pedestrian-A
0.30 %
0.0025 %
Receiver A never reaches 10⁻³
3GPP EPA
0.32 %
0.0025 %
Receiver A near 3 ⋅ 10⁻³ floor
3GPP EVA
13.92 %
0.058 %
Receiver A bounded above 10⁻¹
ITU-R Vehicular-A
21.55 %
0.31 %
Receiver A bounded above 0.2
Numbers at E b /N 0 = 10 dB. Receiver A exhibits the expected unequalized ISI floor; Receiver B with MMSE-LE tracks the AWGN bound to within a few dB. BLER from these BERs depends on the coding stack used in production (we do not model HARQ here).
§4 · Time-varying CSI · what the chain enables
When CSI is noisy, the verified past prior wins.
In any real PHY the channel estimate ĥ t is pilot-budget-limited — the SRS / DMRS density is fixed by the standard, and the per-snapshot estimation noise is substantial relative to per-tap channel coefficients. Below we run an AR(1) drift channel (ρ = 0.97, slow Doppler regime) with realistic per-tap CSI noise (σ ε = 0.18). Three receivers compete at fixed E b /N 0 = 10 dB over 24 channel updates.
h t = ρ·h t−1 + √(1−ρ²)·ν t (slow channel drift) ĥ t = h t + ε t (per-snapshot CSI estimation noise) M̂ t = (1 − w) · ρ · M̂ t−1 + w · ĥ t (recursive smoother — uses sealed past)
Receiver
Avg BER (24 steps)
Worst step
Best step
A — no M t (carrier recovery only)
15.30 %
29.89 %
2.77 %
B — memoryless M t (per-step noisy CSI)
4.67 %
12.70 %
0.39 %
C — M t with verified-past memory hop
2.76 %
7.60 %
0.33 %
The memory-hop receiver wins by 1.69 × on average BER and — more importantly for cell-edge UE behavior — its worst-step BER is roughly halved. Lower worst-step BER directly translates to fewer HARQ retransmissions in the production stack. The trust in the past prior is what §5 secures.
§5 · Transition hash chain · LI & SLA forensics
Each M t is signed by an FPGA-resident PUF key, then chained.
The receiver gain in §3 / §4 is conventional engineering. The reason a cellular operator should care more than a modem vendor is the second product — a per-step signed record of network behavior, in a form regulators and courts accept as evidence. A single bit altered anywhere in the historical channel-state record breaks every downstream signature.
Fig. — 12-step transition chain. One bit modified in M 10 : 10 / 24 links validate (clean baseline = 24 / 24). All H t for t ≥ 10 fail.
σ t = HMAC K PUF (H t−1 ‖ M t )
H t = SHA256(H t−1 ‖ M t ‖ σ t )
verify HMAC K PUF (H t−1 ‖ M t ) ≟ σ t ∀ t
MAC primitive
HMAC-SHA256, 256-bit PUF-derived key
Key custody
AWS F2 FPGA (AFI agfi-085919c09c35982c3 ); 256-bit PUF root never crosses the FPGA boundary
LI compliance
Compatible with ETSI TS 102 232 (LI handover interfaces) — the chain attests when CSI changed, not the content
SLA / roaming evidence
Both parties can verify identical M t chain offline given the public verification key
Tamper sensitivity
1 bit flipped at step k → 100 % of links from k to T reject (demonstrated on T = 24)
§6 · Cell-edge translation
Where the §3 / §4 numbers actually buy something.
Cell-edge UE on 3GPP EVA typically operates with effective E s /N 0 ≈ 0–10 dB after shadowing. The Receiver-A floor in this regime forces lower MCS, more HARQ retransmissions, and higher PUSCH power offsets. A receiver that tracks the AWGN bound to within ~3 dB recovers most of the link budget the cell edge throws away.
MCS index lift
Same channel quality indicator, higher MCS sustainable — equivalent to ~2–3 dB SNR gain on EVA-class channels.
HARQ retransmissions
Lower worst-step BER (§4) directly reduces NACK density at cell edge → less PRB consumed by retransmissions.
Coverage stretch
Effective cell radius scales with link-budget recovered — coverage planning input, not a unit-level claim.
Power overhead
Modal extraction + HMAC sign budget on F2-class FPGA: under measurement; preliminary < 1 ms per snapshot.
§7 · Integration paths
Three places the chain plugs in — pick the one that fits your stack.
L1 / PHY in-line
Modal extraction co-located with channel estimator block; HMAC sign per CSI snapshot. Tightest latency, highest gain. ASIC/FPGA candidate.
CU-DU split (O-RAN)
Chain runs in DU; signed M t records published over E1 / F1 interfaces. Compatible with O-RAN Alliance WG7 specs.
Off-path log signer
PHY untouched; sealed CSI records produced by a side-car appliance from KPI exports. Cleanest deployment, no PHY-cert impact.
Licensing model
Per-cell IP block (FPGA / ASIC), per-DU software license, or hosted seal-verification service. We sign deal structure with the operator's procurement team via the lead financial advisor.
§8 · Honest limitations
What this page does not show.
SISO only. No MIMO, no spatial multiplexing, no beamforming. These would add their own gains and are separate work. The chain primitive is dimension-agnostic; the per-antenna or per-layer M t can be chained the same way.
No OFDM / no HARQ modeled. BER, not BLER, is reported. Production BLER depends on code rate, MCS, and HARQ — none of which are modeled here. The translation from BER to BLER is operator-specific and not claimed.
Perfect CSI in §3. Receiver B in §3 uses the exact channel response. §4 relaxes this with realistic per-snapshot CSI noise; the §4 numbers are the more relevant ones for fielded UE.
Receiver A is single-tap carrier recovery, not state of the art. A real telco modem will run DFE / MLSE / turbo equalization on a competent CSI. This page compares the channel-aware primitive to no channel awareness; comparison against the operator's actual production receiver is the first step of any pilot.
Chain content vs chain attestation. The chain attests which M t was observed and when . It does not encrypt the M t nor attest to UE identity. LI / privacy regimes that require those properties need separate primitives layered on top.
FPGA in this demo is software-emulated for the chain. The 24-step HMAC chain shown in §5 was computed in Python. A separate program runs the same HMAC on a real AWS F2 FPGA AFI; wiring the PHY-side modal extraction to that FPGA is the production integration step.
Sample-size-limited measurements. 200,000 bits per E b /N 0 in §3 caps measurable BER at ~5 × 10⁻⁶; we report "0 errors in N" rather than extrapolating below that.
§9 · For your link-level team
Run it on your own captures.
The Python pipeline that produced every figure on this page is available under NDA. The most useful first step is to repeat §3 on your operator-specific channel soundings, or on captures from your existing UE under test. Modal-extraction parameters and equalizer regularization are exposed; the chain primitive is independent of any PHY choice.
Request engineering briefing Read the whitepaper