The radio now appears as a device in both audio lists — 'Radio (TCI network audio)' — so the receive audio and the voice keyer can both take the CAT link instead of a sound card. No virtual cable, no second card, no Windows mixer between the recording and the air. The transmit side reuses the exchange the tone probe established, with a WAV in place of the sine: the radio asks, we answer with the next slice, and it sets the pace. The message is converted once, up front, rather than per frame — a voice message is a few hundred kilobytes, and resampling inside a callback that has 21 ms to answer would put arithmetic on the path where a late frame is a gap on the air. The PTT is untouched by all this: the keyer keys before and unkeys after exactly as it does with a sound card, so a transmission is bracketed by the same code whichever way the audio travels. And the playback runs OFF the CAT goroutine, since everything else about the rig goes through that one place and a ten-second message would otherwise freeze the frequency display and the antenna following for ten seconds. Two refusals rather than silent failure. A radio that has gone away stops being offered as a device at all, and a radio whose transmit audio source is still the microphone is caught within a fifth of a second — a voice keyer that transmits silence is worse than one that says why it will not.
190 lines
5.6 KiB
Go
190 lines
5.6 KiB
Go
package cat
|
|
|
|
// Playing a recorded message to the radio over TCI — the voice keyer's path.
|
|
//
|
|
// The same exchange the tone probe established, with a WAV in place of the
|
|
// sine: the radio asks for a frame, we answer with the next slice of the
|
|
// message, and it sets the pace. What is added here is the conversion, because
|
|
// a recording is whatever the microphone gave it — 16-bit, often mono, often
|
|
// not 48 kHz — and the radio wants interleaved float32 at the stream's rate.
|
|
//
|
|
// The message is converted ONCE, up front, rather than per frame. A voice
|
|
// message is a few hundred kilobytes; resampling it inside the callback would
|
|
// put arithmetic on the path that has 21 ms to answer, and a late frame is a
|
|
// gap in what goes out.
|
|
|
|
import (
|
|
"encoding/binary"
|
|
"fmt"
|
|
"math"
|
|
"time"
|
|
)
|
|
|
|
// tciTXFirstAskTimeout is how long to wait for the radio to ask for the first
|
|
// frame before giving up.
|
|
//
|
|
// It answers within a frame or two when it is going to answer at all, so this
|
|
// is generous. When it stays quiet the cause is always the same — the transmit
|
|
// audio source is the microphone rather than TCI — and a fifth of a second of
|
|
// carrier is a cheap way to find that out.
|
|
const tciTXFirstAskTimeout = 200 * time.Millisecond
|
|
|
|
// PlayTXAudio sends one message and returns when it has all been handed over,
|
|
// or when stop is closed.
|
|
//
|
|
// The PTT is NOT touched here. The voice keyer keys before calling and unkeys
|
|
// after, exactly as it does with a sound card, so the transmission is bracketed
|
|
// by the same code whichever way the audio travels.
|
|
func (t *TCI) PlayTXAudio(pcm []byte, rate, ch, bits int, stop <-chan struct{}) error {
|
|
t.mu.Lock()
|
|
connected := t.conn != nil
|
|
t.mu.Unlock()
|
|
if !connected {
|
|
return fmt.Errorf("not connected to the radio")
|
|
}
|
|
|
|
t.audio.mu.Lock()
|
|
outRate := t.audio.rate
|
|
t.audio.mu.Unlock()
|
|
if outRate <= 0 {
|
|
outRate = 48000
|
|
}
|
|
|
|
mono := decodeToMono(pcm, ch, bits)
|
|
if len(mono) == 0 {
|
|
return fmt.Errorf("the message is empty")
|
|
}
|
|
if rate > 0 && rate != outRate {
|
|
mono = resampleLinear(mono, rate, outRate)
|
|
}
|
|
|
|
// Served from here on. The callback does nothing but copy and interleave,
|
|
// which is what keeps it inside the frame interval.
|
|
pos := 0
|
|
done := make(chan struct{})
|
|
var closed bool
|
|
t.setTXFeed(func(samples int) []byte {
|
|
if samples <= 0 {
|
|
samples = 2048
|
|
}
|
|
pairs := samples / 2
|
|
if pos >= len(mono) {
|
|
if !closed {
|
|
closed = true
|
|
close(done)
|
|
}
|
|
return nil
|
|
}
|
|
payload := make([]byte, samples*4)
|
|
le := binary.LittleEndian
|
|
for i := 0; i < pairs; i++ {
|
|
var v float32
|
|
if pos < len(mono) {
|
|
v = mono[pos]
|
|
pos++
|
|
}
|
|
bits := math.Float32bits(v)
|
|
le.PutUint32(payload[(i*2)*4:], bits) // left
|
|
le.PutUint32(payload[(i*2+1)*4:], bits) // right
|
|
}
|
|
return payload
|
|
})
|
|
defer t.setTXFeed(nil)
|
|
|
|
// Nothing asked for in a fifth of a second means nothing is listening.
|
|
// Reported plainly: the message would otherwise go out as silence, and a
|
|
// voice keyer that transmits silence is worse than one that refuses.
|
|
deadline := time.Now().Add(tciTXFirstAskTimeout)
|
|
for time.Now().Before(deadline) {
|
|
t.audio.mu.Lock()
|
|
asked := t.audio.txSent > 0
|
|
t.audio.mu.Unlock()
|
|
if asked {
|
|
break
|
|
}
|
|
select {
|
|
case <-stop:
|
|
return nil
|
|
case <-time.After(10 * time.Millisecond):
|
|
}
|
|
}
|
|
t.audio.mu.Lock()
|
|
asked := t.audio.txSent
|
|
t.audio.mu.Unlock()
|
|
if asked == 0 {
|
|
return fmt.Errorf("the radio did not ask for any audio — set its transmit audio source to TCI instead of the microphone")
|
|
}
|
|
|
|
// The radio drains the message at real time, so this waits for the feed to
|
|
// run out. The cap is the message's own length with a second to spare: a
|
|
// radio that stops asking mid-message must not hold the transmitter up.
|
|
limit := time.Duration(float64(len(mono))/float64(outRate)*float64(time.Second)) + time.Second
|
|
select {
|
|
case <-done:
|
|
case <-stop:
|
|
case <-time.After(limit):
|
|
debugLog.Printf("TCI: the radio stopped asking for audio before the message ended")
|
|
}
|
|
return nil
|
|
}
|
|
|
|
// decodeToMono turns interleaved PCM into one channel of -1…1 floats.
|
|
func decodeToMono(pcm []byte, ch, bits int) []float32 {
|
|
if ch <= 0 {
|
|
ch = 1
|
|
}
|
|
switch bits {
|
|
case 16:
|
|
frame := ch * 2
|
|
out := make([]float32, 0, len(pcm)/frame+1)
|
|
for i := 0; i+frame <= len(pcm); i += frame {
|
|
var sum float32
|
|
for c := 0; c < ch; c++ {
|
|
v := int16(uint16(pcm[i+c*2]) | uint16(pcm[i+c*2+1])<<8)
|
|
sum += float32(v) / 32768
|
|
}
|
|
out = append(out, sum/float32(ch))
|
|
}
|
|
return out
|
|
case 8:
|
|
// Unsigned, centred on 128 — the one format where silence is not zero.
|
|
out := make([]float32, 0, len(pcm)/ch+1)
|
|
for i := 0; i+ch <= len(pcm); i += ch {
|
|
var sum float32
|
|
for c := 0; c < ch; c++ {
|
|
sum += (float32(pcm[i+c]) - 128) / 128
|
|
}
|
|
out = append(out, sum/float32(ch))
|
|
}
|
|
return out
|
|
}
|
|
return nil
|
|
}
|
|
|
|
// resampleLinear moves samples from one rate to another.
|
|
//
|
|
// Linear interpolation, which is crude and entirely adequate here: a voice
|
|
// recording at 16 kHz going to 48 kHz is being INTERPOLATED, and interpolation
|
|
// invents no frequencies to alias. Going the other way would want a filter
|
|
// first, but a message recorded above the radio's stream rate is not a case
|
|
// that arises — the recorder works at 16 kHz and radios stream at 48.
|
|
func resampleLinear(in []float32, from, to int) []float32 {
|
|
if from <= 0 || to <= 0 || from == to || len(in) == 0 {
|
|
return in
|
|
}
|
|
ratio := float64(from) / float64(to)
|
|
n := int(float64(len(in)) / ratio)
|
|
out := make([]float32, n)
|
|
for i := 0; i < n; i++ {
|
|
src := float64(i) * ratio
|
|
j := int(src)
|
|
frac := float32(src - float64(j))
|
|
if j+1 < len(in) {
|
|
out[i] = in[j]*(1-frac) + in[j+1]*frac
|
|
} else {
|
|
out[i] = in[len(in)-1]
|
|
}
|
|
}
|
|
return out
|
|
}
|