M6 test realism — a moving-UE handover bench (conducted path-loss matrix, then an over-the-air walk)

Aberta
#563 1 comentário 0 reações 0 responsáveis Ver no GitHub

Mantenedores costumam responder em até 1 dia

Ninguém assumiu esta issue ainda.

Avaliação

Dificuldade
5/5
Tempo estimado
Mais de uma semana
Facilidade para iniciantes
25/100
Tipo de issue
Funcionalidade
Clareza
Razoavelmente clara
Status de atividade
Ativa
Stack de tecnologia
c, shell

Direção de pesquisa

Start with the “An emulated crossing” and “Not yet” sections of docs/cell-m6.md, then review scripts/radio_antenna_check.sh for the existing calibration approach. The first hardware check is to measure RSSI across every cell/UE pair with the matrix at maximum loss; use those results to assess isolation and the need for shielding or different placement. Done means the matrix supports calibrated, repeatable multi-UE M6 regression runs and one recorded over-the-air walk meeting the listed acceptance criteria.

Escrita pelo modelo de indexação a partir do texto da issue.

Descrição

enhancement

Part of the 5G-NR RAN EPIC #254. Follows M6 #260: every M6 mechanism (CHO, DAPS on one channel and across channels, slot migration, per-cell keys, sequence handling and handover state cleanup) is merged and passes on the rig. What is not covered is test realism: every crossing so far is emulated.

What the current emulation does not test

As recorded in docs/cell-m6.md ("An emulated crossing", "Not yet"):

  • Only the downlink moves. CELL_DU_TX_RAMP ramps the DUs' TX power against each other. The UE's uplink path loss to each cell stays constant, so the DUs' uplink measurements, uplink DAPS and the target hearing the UE never see a real fade.
  • One UE and one speed. The ramp is one setting, about 2 dB/s of margin. There is no crowd: no concurrent handovers, no target admitting under load, no UE-to-UE interference.
  • No fading. The ramps are piecewise linear. A3 hysteresis and time-to-trigger have never met fast fading, so ping-pong under fading is untested; only a selftest model covers it.
  • Range. The DUs' ~20 dB TX range can't take a UE all the way round 1 → 2 → 3 → 1. The way back into cell 1 is out of reach.
  • Calibration. The rig's power-to-RSSI relation isn't linear (−8 dB of offset reads about −9 dB, −9 dB about −12 dB; cause not established). The tests use measured margins because the injected loss isn't known accurately.

Goal

A bench where both link directions between every UE and every cell follow a scripted trajectory: real path loss, optionally fading, several UEs at once. It should be repeatable enough to run as a regression, with the injected loss known to about 1 dB.

Options

A. Conducted path-loss matrix (recommended first)

Coax from each radio's antenna ports through programmable attenuators into a mesh, so every cell↔UE link has its own controlled loss in both directions.

  • Size today: 3 cells (one 8852CE per J6412) and up to 3 UEs (the second 8852CE per box) gives a 3×3 cell↔UE mesh, plus UE↔UE and cell↔cell paths if those should be controlled too. A 4×4 or 6-port full-mesh attenuator unit leaves room for a 4th cell or more UEs.
  • Per-port range: ≥60 dB in ≤1 dB steps, settable at ≥10 Hz from a script, so that steps don't read as cliffs to the L3 filter. A faster update rate would also allow a slow fading profile, below.
  • Two chains per card: cable both through a 2:1 combiner per radio, or cable one chain and terminate the other (50 Ω) with the TX/RX path mask on that chain. That's a decision to make, not a default.
  • Isolation is the hard part on this rig. Each J6412 holds a cell radio and a UE radio a few centimetres apart. Leakage that bypasses the matrix (board-level coupling, pigtails, enclosure openings) sets the floor of the usable range, and that floor is unknown. Measure it first: with the matrix at maximum loss, read RSSI across every pair. The result decides whether enclosures or shielding are needed, or a different cell/UE placement across boxes.
  • What it doesn't give: multipath, delay spread and Doppler. At walking or robot speed, 5 GHz indoors, delay spread sits inside the OFDM guard interval and Doppler is a few Hz, so path loss plus slow fading is most of what matters for handover. A true channel emulator is option C.
B. Over-the-air with a moving UE

A UE on a cart, robot or person, walking past cells spread across rooms.

  • Needs a portable UE node: a host with an M.2 E-key 8852CE (or another radio whose AdapterCaps passes the cell UE's check: tsf_write_ok, tbtt_follows_tsf, beacon_backoff_ctl, hw_beacon_txtsf), on battery. It must pass box_acceptance.sh like any rig box. Whether a USB Kestrel (8852CU) passes the UE caps hasn't been checked.
  • Clocks: the UE syncs over the air (TsfSync) and needs no PTP. Most KPIs come from the PTP-locked DUs and the CU; the UE's interruption and delivery timings are on its own monotonic clock.
  • Trade-off: real fading and real geometry, but runs are not repeatable. Good for occasional validation, poor as a regression.
C. Channel emulator (later, if A shows a need)

A commercial MIMO channel emulator, for multipath and Doppler profiles. Expensive. Only worth it if conducted runs leave a question that path loss plus slow fading can't answer.

Recommendation: A as the regression bench. B once, as an end-to-end sanity walk after A is in place. C only on evidence.

Software (devourer + testrig)

  • testrig scripts/m6_mobility.sh: drives the attenuators from a scenario file (UE trajectories → per-link loss from a simple propagation model, with optional fading). It runs on the n300-1 CU host and logs every attenuator step as JSONL on the PTP timebase, so KPIs line up with the cell logs.
  • Calibration step: sweep each path and record attenuation against RSSI at both ends (scripts/radio_antenna_check.sh style). This replaces "measured margins" with known loss, and should settle or explain the non-linear power-to-RSSI relation.
  • Retire CELL_DU_TX_RAMP / CELL_ATTEN_* from the M6 suite once the matrix covers the same cases, or keep them as a quick no-hardware fallback.
  • Multi-UE scenarios: UEs moving together (a convoy, so concurrent handovers) and crossing each other.

Acceptance (KPIs reported as distributions, per the epic)

  • Isolation floor measured per path. Usable range ≥40 dB of controlled differential between any two cells at one UE.
  • Calibration: injected loss against measured RSSI, per path, linear to within ±1 dB over the usable range, or documented where not.
  • The M6 suite on the matrix, uplink and downlink both following the trajectory: handover, DAPS (same channel, across channels, source failure), crossing 1 → 2 → 3 → 1 (the full tour).
  • A3 under fading: ping-pong rate and trigger delay across speed × hysteresis × time-to-trigger, against the selftest model.
  • Several UEs: ≥3 UEs handing over within one window. The target admits and serves them all, with interruption and loss per UE.
  • One over-the-air walk (B) with the same KPIs, recorded.

Open decisions

  • Matrix size and vendor (budget); one chain or two per radio.
  • Whether isolation forces enclosures, or moving UE radios out of the cell boxes (more hosts).
  • More UEs: more J6412s with 8852CEs, or USB radios, if one passes the UE caps.
  • Portable UE host for B.

Relationships

#260 (M6) leaves test realism open; this issue picks it up. #457 (multi-receiver ground station) would reuse the same attenuator matrix for its N-transmitter case.

Linguagem predominante
C
Estrelas
56
Forks
29
Merge médio
3h 43min
PRs com merge (30d)
75

Preparar o ambiente

Este projeto não oferece contêiner de desenvolvimento, Dockerfile nem guia de contribuição, então a configuração fica por sua conta: comece pelo README e veja nosso guia da primeira contribuição para os passos gerais.

Primeiros passos

  1. Leia a issue inteira e depois o guia de contribuição do projeto.
  2. Comente na issue dizendo que vai assumir — evita que duas pessoas façam o mesmo trabalho.
  3. Faça um fork do repositório e trabalhe em uma branch.
  4. Abra um pull request que referencie o número da issue.

Mais de OpenIPC/devourer

Todas as issues de OpenIPC/devourer

Issues semelhantes

Mais issues de C

Receba novas issues na sua caixa de entrada

Um resumo curto de issues do GitHub para quem está começando.