laforge has submitted this change. ( https://gerrit.osmocom.org/c/osmo-e1-hardware/+/29609 )

Change subject: icE1usb fw/gpsdo: Use PID loop instead of ad-hoc algo ......................................................................

icE1usb fw/gpsdo: Use PID loop instead of ad-hoc algo

The original algo is just somehting I originally came up with on the spot and worked and I never got the time to revisit.

Now after some testing, I implemented a PID loop that seems to present faster lock time and at least as good stability.

Parameters were not thorougly optimized just some 'hand' tuning trying to simulate the behavior.

Signed-off-by: Sylvain Munaut tnt@246tNt.com Change-Id: Ie3ea7243aa4234585f1ace222632bb5580caca75 --- M firmware/ice40-riscv/icE1usb/gpsdo.c 1 file changed, 22 insertions(+), 28 deletions(-)

Approvals: Jenkins Builder: Verified laforge: Looks good to me, approved

diff --git a/firmware/ice40-riscv/icE1usb/gpsdo.c b/firmware/ice40-riscv/icE1usb/gpsdo.c index 5975100..573c0bb 100644 --- a/firmware/ice40-riscv/icE1usb/gpsdo.c +++ b/firmware/ice40-riscv/icE1usb/gpsdo.c @@ -49,8 +49,8 @@

/* Fine tracking */ struct { - int acc; - int div; + int acc; /* Accumulated error */ + int integral; /* PID Integral term */ } fine;

} g_gpsdo; @@ -60,6 +60,7 @@ * VCXTO parameters * * - iKv is reciprocal sensitivity vs 'coarse' count for fast initial acquisition + * - Kp, Ki, Kd are params for the fine-tracking PID loop * * Note that the spec if often a guaranteed minimum range and goes * from ~0.1V to 3.2V instead of 0-3.3V so actual sensitivity is @@ -67,13 +68,22 @@ */ static const struct { int iKv; /* hi-count / Hz (.8 fixed point) */ + int Kp; /* PID proportional term (.12 fixed point) */ + int Ki; /* PID integral term (.12 fixed point) */ + int Kd; /* PID differential term (.12 fixed point) */ } vctxo_params[] = { [VCTXO_TAITIEN_VT40] = { .iKv = 300, /* +- 50 ppm pull range => ~ 0.75 Hz / hi-count (set to 0.85) */ + .Kp = 14336, + .Ki = 410, + .Kd = -6144, },

[VCTXO_SITIME_SIT3808_E] = { .iKv = 160, /* +- 100 ppm pull range => ~ 1.50 Hz / hi-count (set to 1.6) */ + .Kp = 7168, + .Ki = 205, + .Kd = -4096, }, };

@@ -188,7 +198,7 @@

/* Reset the long term error tracking */ g_gpsdo.fine.acc = 0; - g_gpsdo.fine.div = 0; + g_gpsdo.fine.integral = 0; }

static void @@ -219,7 +229,6 @@ _gpsdo_fine_track(uint32_t tick_diff) { int freq_diff = (int)tick_diff - TARGET; - uint16_t tune;

/* Did we deviate too much ? */ if ((freq_diff < -2*MAX_DEV_FINE) || (freq_diff > 2*MAX_DEV_FINE)) { @@ -227,30 +236,16 @@ return; }

- /* Accumulate long term error */ + /* Accumulate long term error and integrate it */ g_gpsdo.fine.acc += freq_diff; + g_gpsdo.fine.integral += g_gpsdo.fine.acc;

- /* Update fine tuning value */ - if (((g_gpsdo.fine.acc >= 0) && (freq_diff > 0)) || - ((g_gpsdo.fine.acc <= 0) && (freq_diff < 0))) - { - g_gpsdo.tune.fine -= freq_diff; - } - - if (g_gpsdo.fine.acc) { - if (++g_gpsdo.fine.div > 10) { - g_gpsdo.fine.div = 0; - if (g_gpsdo.fine.acc > 0) - g_gpsdo.tune.fine--; - else if (g_gpsdo.fine.acc < 0) - g_gpsdo.tune.fine++; - } - } else { - g_gpsdo.fine.div = 0; - } - - /* Compute value with a bias from long term accumulator */ - tune = g_gpsdo.tune.fine - (g_gpsdo.fine.acc / 2); + /* PID */ + g_gpsdo.tune.fine = 2048 - (( + vctxo_params[g_gpsdo.vctxo].Kp * g_gpsdo.fine.acc + + vctxo_params[g_gpsdo.vctxo].Ki * g_gpsdo.fine.integral + + vctxo_params[g_gpsdo.vctxo].Kd * freq_diff + ) >> 12);

/* If fine tune is getting close to boundary, do our * best to transfer part of it to coarse tuning */ @@ -260,13 +255,12 @@

g_gpsdo.tune.coarse += coarse_adj; g_gpsdo.tune.fine -= coarse_adj << 6; - tune -= coarse_adj << 6;

pdm_set(PDM_CLK_HI, true, g_gpsdo.tune.coarse, false); }

/* Apply fine */ - pdm_set(PDM_CLK_LO, true, tune, false); + pdm_set(PDM_CLK_LO, true, g_gpsdo.tune.fine, false);

/* Debug */ #ifdef GPSDO_DEBUG