SQ Stereo/Quadraphonic decoder with shadow vector DES as LADSPA plugin.
"hilbert_stereo_####taps" plugins are modifications of SWH's "hilbert_1440" plugin, with more filter taps and two input channels. Those plugins are responsible for 90-degrees phase shift necessary for SQ-decoding plugins.
"hilbert_stereo_####taps_soxhack" plugins are workaround over SoX v14.4.2 limitation with ladspa plugins: it requires that plugins have same amount of outputs as inputs.
"sq_decoder_nologic" plugin is naive reference implementation for SQ-decoder. It has no directional enhancements and is rough equivalent of early "no-logic" decoders.
"sq_decoder_shadow_vector" is implementation of SQ directional enhancement system "Decoder for quadraphonic playback" patented by Lynn T. Olson in 1977. It seems by vague internet sources, that this system was never actually produced in seventies. This method seems to be extremely sensitive to phase-shift network quality, specially in left-center right-center separation. Phase-shifting with analogue components is complex (pun intended:) problem and can never reach accuracy which is easily available with now-days digital methods and equipment.
One way to use those plugins is with ecasound:
LADSPA_PATH=./ladspa-sq-sv ecasound -x -i sq-test-tones.flac
-el:hilbert_stereo_200taps -el:sq_decoder_shadow_vector -f:,4 -o test.wav
Example usage with SoX:
with audio interface with 4 input channels:
LADSPA_PATH=./ladspa-sq-sv sox -r 192k -t alsa hw:1 -t alsa hw:0
ladspa -l hilbert_stereo_800taps_soxhack ladspa -l sq_decoder_shadow_vector
with sound card with 2 input channels:
LADSPA_PATH=./ladspa-sq-sv sox -r 48k -t alsa hw:1 -t alsa hw:0
remix -m 1 2 0 0 ladspa -l hilbert_stereo_200taps_soxhack ladspa -l sq_decoder_shadow_vector
In general, feed encoded signal to hilbert-plugins, and hilbert_plugin output to sq-decoder-plugin.
My setup consists of behringer u-phoria umc404hd audio interface and intel nuc minicomputer. Turntable is connected to audio interface and minicomputer feeds amplifier via hdmi connector. I experimented with raspberry pi3, it runs fine with 48k samplerate but not with 96k nor 192k rates.
I'm quite happy with plugins current performance. I's sounds way better then those hardware SQ decoders that i have listened: sony sqd-2020 and sqd-2050. It passes test signals flawlessly. Directional separation for 8 cardinal directions is infinite. :)
This is for linux. My windows-programming skills are somewhat rusty for now.. It's should be trivial to port to windows and mac by someone who knows how develop to those platforms.
Turns out that pulseaudio emulation over pipewire works better than direct ALSA. Pipewire cleverly resamples its streams as needed to avoid buffer underruns caused by slightly different clock speeds on different sound cards. Usage:
sox -q -r 192k -b 32 -t pulseaudio behringer_ai_stereo_source -t pulseaudio behringer_ai_quadro_sink remix 1v1 2v1 0 0 ladspa -l hilbert_stereo_2000taps_soxhack ladspa -l sq_decoder_shadow_vector
Uses about 40% of the one CPU on N97-equpped minicomputer. Honestly, there is no need to run it with those maxed-out parameters. 48k sample rate and 200 taps phase-shift-filter sounds the same for me. CPU usage drops to a few % of one CPU. My setup still uses maxed out config, because ...!
About quality: I have re-capped the Sony SQD-2020 decoder and my digital decoder built around the Behringer UMC404HD audio interface. Digital decoder sounds much cleaner, without steering artefacts so obvious on Sony. Also, Sony adds some amount of hiss to the sound. Yet the Sony decoder sees a bit more use, mostly because illuminated dancing needles are so cool!
I hope that the information here helps you to build your own SQ decoder and to enjoy an obscure wierd part of the history of sound recording.
-- With regards, Markko Merzin markko.merzin@gmail.com
