SDD V1.2 · 2026-08-04

MRRC-FT8
Software Design Description

Canonical design record for the headless FT8/FT4 server — architecture, interfaces, operational model, and runtime evolution.

Browse SDD Chapters Back to Product Page
Overview

SDD Scope

IBM TeamSD aligned, 15 chapters, implementation-grounded content.

AttributeValue
Document IDSDD-MRRC-FT8-2026-001
VersionV1.2
Baseline Date2026-08-04
PhaseLive on radio.vlsc.net:9988 (FT-710)
RuntimeWSJT-X 3.0.2 Improved DSP + FastAPI + Uvicorn + NumPy
FrontendHTML5, CSS3, vanilla JavaScript, Canvas PWA
Core InterfacesFastAPI REST/WS + supervised DSP Worker IPC + rigctld loopback
Latest Update

What Landed In V1.2

FT-710 Rig Levels via Raw CAT

ATT / PREAMP / AGC / RF gain now switch through raw \send_raw CAT frames instead of hamlib's L/l level path, which the FT-710 never answers (verified live 2026-08-04 against the station rigctld). Codes: RA01;=6 dB, PA01;=10 dB, GT06;=AUTO — hamlib 4.7.2 sends the wrong GT04; — and RG0<NNN>;=0..1. The drawer reads the rig's real state and drops the stale snapshot after every write.

FT-710 Filter Width Raw-SH

Hamlib 4.6.2 mis-framed the width command (SH0NN; instead of SH00NN;) so M <mode> <width> never changed the rig. The width now goes through raw \send_raw with the true SH-register readback (1.8 / 2.4 / 3.0 kHz), compatible with the hamlib 4.7.2 upgrade. TX power (PC0…;) is rejected by the FT-710 and is deliberately not exposed.

MRRC-FT8 web cockpit — live FT8 band activity and decodes on 14.074 MHz (20 m)

The live cockpit: 20 m FT8 band activity and decodes on 14.074 MHz.

V1.1 Baseline

What Landed In V1.1

UtcRing Eviction Root Cause

Physical ring positions were keyed to (absolute − base) % capacity, so once eviction advanced base, reads addressed samples with the new base while data never moved — a shift growing 1:1 with wall clock, wrapping at capacity. Positions now key to the absolute sample index alone (X % capacity): eviction never moves data, base is purely a bounds check.

Replay Opposite-TX-Slot Phase

The reply path was missing the UC-003 "opposite TX slot": TxDriver gated on a hard-coded even parity that never changed, so replies to partners on the same phase always collided. The sequencer now carries tx_phase = 1 − slot_id % 2 and the QSO alternates to completion.

Manual Reply Decision Window

Replies always missed the slot right after the message they answer. The TX decision is now provisional: the window stays open until decision_cutoff with polling instead of sleeping, and a fit guard defers past-fit-deadline replies to the next eligible slot.

3 kHz Waterfall + Band Selector

Spectrum frames now emit only the 3 kHz FT8 passband — no blank half-canvas, half the payload. A band selector in the cockpit top bar tunes 7/14/21/28 MHz FT8 dial frequencies via the lease-gated /radio/band mutation.

Chapter Index

15 SDD Chapters

01 Executive Summary

02 Business Direction

03 Project Definition

04 System Context

05 Non-Functional Requirements

06 Use Case Model

07 Subject Area Model

08 Architecture Decisions

09 Architecture Overview

10 Service Model

11 Component Model

12 Operational Model

13 Feasibility Assessment

14 Version History

15 PTT Safety Architecture

Architecture Decisions

AD-001 ... AD-015

AD-001 Headless Layered Server AD-002 WSJT-X C ABI AD-003 Supervised DSP Worker AD-004 Fixed Audio Rate Domains AD-005 Single DSP Owner + Lock AD-006 UTC Epoch Slot Discipline AD-007 Sequencer/PTT Safety AD-008 rigctld Serial Owner AD-009 Caddy Edge + Cookie Auth AD-010 Single Control Lease AD-011 Landscape-First Cockpit AD-012 Human Selection, Single QSO AD-013 Separate Bounded Streams AD-014 SQLite + ADIF AD-015 Native Service Deployment