osmo-bts/src/common/power_control.c

310 lines
10 KiB
C
Raw Normal View History

/* MS Power Control Loop L1 */
/* (C) 2014 by Holger Hans Peter Freyther
power_control: implement EWMA based Uplink power filtering So far the Uplink power control loop did not filter the Uplink RSSI measurements (reported by the BTS) at all. The lack of filtering makes our implementation too quick on the trigger, so in the real deployments there will be unneeded Tx power oscillations. In order to reduce this effect, let's implement a very simple EWMA (also known as Single Pole IIR) filtering that is defined as follows: Avg[n] = a * Pwr[n] + (1 - a) * Avg[n - 1] where parameter 'a' determines how much weight of the latest UL RSSI measurement result 'Pwr[n]' carries vs the weight of the average 'Avg[n - 1]'. The value of 'a' is usually a float in range 0 .. 1, so: - value 0.5 gives equal weight to both 'Pwr[n]' and 'Avg[n - 1]'; - value 1.0 means no filtering at all (pass through); - value 0.0 makes no sense. This formula was further optimized with the use of '+=' operator. The floating point math was also eliminated by scaling everything up (by 100). For more details, see: https://en.wikipedia.org/wiki/Moving_average https://en.wikipedia.org/wiki/Low-pass_filter#Simple_infinite_impulse_response_filter https://tomroelandts.com/articles/low-pass-single-pole-iir-filter The EWMA filtering is now *enabled by default*, but can be disabled or (re-)configured over the VTY at any time: ! Completely disable filtering no uplink-power-filtering ! Enable EWMA smoothing with the given parameters uplink-power-filtering algo ewma beta <1-99> Note that the VTY command expects 'beta' instead of 'alpha': alpha = (100 - beta) and the value must be in %. This is done for simplicity: 1% means lowest smoothing, 99% means highest smoothing. Let's say we have EWMA filtering enabled with alpha = 0.4, and get -98 dBm on the input, while the last output value was -60 dBm. The new output would be: Avg[n] = 0.4 * Pwr[n] + 0.6 * Avg[n - 1] Avg[n] = (0.4 * -98) + (0.6 * -60) Avg[n] = -75.2 => around -75 Of course, this is not a silver bullet, but better than nothing. Change-Id: Ib6dcadbf14ef59696c6a546bd323bda92d399f17 Related: SYS#4916
2020-10-13 14:40:24 +00:00
* Contributions by sysmocom - s.m.f.c. GmbH <info@sysmocom.de>
*
* All Rights Reserved
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU Affero General Public License as published by
* the Free Software Foundation; either version 3 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU Affero General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*
*/
#include <stdint.h>
#include <unistd.h>
#include <errno.h>
#include <inttypes.h>
#include <osmo-bts/logging.h>
#include <osmo-bts/bts.h>
#include <osmo-bts/gsm_data.h>
#include <osmo-bts/measurement.h>
#include <osmo-bts/bts_model.h>
#include <osmo-bts/l1sap.h>
#include <osmo-bts/power_control.h>
power_control: implement EWMA based Uplink power filtering So far the Uplink power control loop did not filter the Uplink RSSI measurements (reported by the BTS) at all. The lack of filtering makes our implementation too quick on the trigger, so in the real deployments there will be unneeded Tx power oscillations. In order to reduce this effect, let's implement a very simple EWMA (also known as Single Pole IIR) filtering that is defined as follows: Avg[n] = a * Pwr[n] + (1 - a) * Avg[n - 1] where parameter 'a' determines how much weight of the latest UL RSSI measurement result 'Pwr[n]' carries vs the weight of the average 'Avg[n - 1]'. The value of 'a' is usually a float in range 0 .. 1, so: - value 0.5 gives equal weight to both 'Pwr[n]' and 'Avg[n - 1]'; - value 1.0 means no filtering at all (pass through); - value 0.0 makes no sense. This formula was further optimized with the use of '+=' operator. The floating point math was also eliminated by scaling everything up (by 100). For more details, see: https://en.wikipedia.org/wiki/Moving_average https://en.wikipedia.org/wiki/Low-pass_filter#Simple_infinite_impulse_response_filter https://tomroelandts.com/articles/low-pass-single-pole-iir-filter The EWMA filtering is now *enabled by default*, but can be disabled or (re-)configured over the VTY at any time: ! Completely disable filtering no uplink-power-filtering ! Enable EWMA smoothing with the given parameters uplink-power-filtering algo ewma beta <1-99> Note that the VTY command expects 'beta' instead of 'alpha': alpha = (100 - beta) and the value must be in %. This is done for simplicity: 1% means lowest smoothing, 99% means highest smoothing. Let's say we have EWMA filtering enabled with alpha = 0.4, and get -98 dBm on the input, while the last output value was -60 dBm. The new output would be: Avg[n] = 0.4 * Pwr[n] + 0.6 * Avg[n - 1] Avg[n] = (0.4 * -98) + (0.6 * -60) Avg[n] = -75.2 => around -75 Of course, this is not a silver bullet, but better than nothing. Change-Id: Ib6dcadbf14ef59696c6a546bd323bda92d399f17 Related: SYS#4916
2020-10-13 14:40:24 +00:00
/* We don't want to deal with floating point, so we scale up */
#define EWMA_SCALE_FACTOR 100
/* Base Low-Pass Single-Pole IIR Filter (EWMA) formula:
*
* Avg[n] = a * Pwr[n] + (1 - a) * Avg[n - 1]
*
* where parameter 'a' determines how much weight of the latest UL RSSI measurement
* result 'Pwr[n]' carries vs the weight of the average 'Avg[n - 1]'. The value of
* 'a' is usually a float in range 0 .. 1, so:
*
* - value 0.5 gives equal weight to both 'Pwr[n]' and 'Avg[n - 1]';
* - value 1.0 means no filtering at all (pass through);
* - value 0.0 makes no sense.
*
* Further optimization:
*
* Avg[n] = a * Pwr[n] + Avg[n - 1] - a * Avg[n - 1]
* ^^^^^^ ^^^^^^^^^^
*
* a) this can be implemented in C using '+=' operator:
*
* Avg += a * Pwr - a * Avg
* Avg += a * (Pwr - Avg)
*
* b) everything is scaled up by 100 to avoid floating point stuff:
*
* Avg100 += A * (Pwr - Avg)
*
* where 'Avg100' is 'Avg * 100' and 'A' is 'a * 100'.
*
* For more details, see:
*
* https://en.wikipedia.org/wiki/Moving_average
* https://en.wikipedia.org/wiki/Low-pass_filter#Simple_infinite_impulse_response_filter
* https://tomroelandts.com/articles/low-pass-single-pole-iir-filter
*/
static int8_t do_pf_ewma(const struct bts_power_ctrl_params *params,
struct lchan_power_ctrl_state *state,
const int8_t Pwr)
power_control: implement EWMA based Uplink power filtering So far the Uplink power control loop did not filter the Uplink RSSI measurements (reported by the BTS) at all. The lack of filtering makes our implementation too quick on the trigger, so in the real deployments there will be unneeded Tx power oscillations. In order to reduce this effect, let's implement a very simple EWMA (also known as Single Pole IIR) filtering that is defined as follows: Avg[n] = a * Pwr[n] + (1 - a) * Avg[n - 1] where parameter 'a' determines how much weight of the latest UL RSSI measurement result 'Pwr[n]' carries vs the weight of the average 'Avg[n - 1]'. The value of 'a' is usually a float in range 0 .. 1, so: - value 0.5 gives equal weight to both 'Pwr[n]' and 'Avg[n - 1]'; - value 1.0 means no filtering at all (pass through); - value 0.0 makes no sense. This formula was further optimized with the use of '+=' operator. The floating point math was also eliminated by scaling everything up (by 100). For more details, see: https://en.wikipedia.org/wiki/Moving_average https://en.wikipedia.org/wiki/Low-pass_filter#Simple_infinite_impulse_response_filter https://tomroelandts.com/articles/low-pass-single-pole-iir-filter The EWMA filtering is now *enabled by default*, but can be disabled or (re-)configured over the VTY at any time: ! Completely disable filtering no uplink-power-filtering ! Enable EWMA smoothing with the given parameters uplink-power-filtering algo ewma beta <1-99> Note that the VTY command expects 'beta' instead of 'alpha': alpha = (100 - beta) and the value must be in %. This is done for simplicity: 1% means lowest smoothing, 99% means highest smoothing. Let's say we have EWMA filtering enabled with alpha = 0.4, and get -98 dBm on the input, while the last output value was -60 dBm. The new output would be: Avg[n] = 0.4 * Pwr[n] + 0.6 * Avg[n - 1] Avg[n] = (0.4 * -98) + (0.6 * -60) Avg[n] = -75.2 => around -75 Of course, this is not a silver bullet, but better than nothing. Change-Id: Ib6dcadbf14ef59696c6a546bd323bda92d399f17 Related: SYS#4916
2020-10-13 14:40:24 +00:00
{
const uint8_t A = params->pf.ewma.alpha;
int *Avg100 = &state->avg100_rxlev_dbm;
power_control: implement EWMA based Uplink power filtering So far the Uplink power control loop did not filter the Uplink RSSI measurements (reported by the BTS) at all. The lack of filtering makes our implementation too quick on the trigger, so in the real deployments there will be unneeded Tx power oscillations. In order to reduce this effect, let's implement a very simple EWMA (also known as Single Pole IIR) filtering that is defined as follows: Avg[n] = a * Pwr[n] + (1 - a) * Avg[n - 1] where parameter 'a' determines how much weight of the latest UL RSSI measurement result 'Pwr[n]' carries vs the weight of the average 'Avg[n - 1]'. The value of 'a' is usually a float in range 0 .. 1, so: - value 0.5 gives equal weight to both 'Pwr[n]' and 'Avg[n - 1]'; - value 1.0 means no filtering at all (pass through); - value 0.0 makes no sense. This formula was further optimized with the use of '+=' operator. The floating point math was also eliminated by scaling everything up (by 100). For more details, see: https://en.wikipedia.org/wiki/Moving_average https://en.wikipedia.org/wiki/Low-pass_filter#Simple_infinite_impulse_response_filter https://tomroelandts.com/articles/low-pass-single-pole-iir-filter The EWMA filtering is now *enabled by default*, but can be disabled or (re-)configured over the VTY at any time: ! Completely disable filtering no uplink-power-filtering ! Enable EWMA smoothing with the given parameters uplink-power-filtering algo ewma beta <1-99> Note that the VTY command expects 'beta' instead of 'alpha': alpha = (100 - beta) and the value must be in %. This is done for simplicity: 1% means lowest smoothing, 99% means highest smoothing. Let's say we have EWMA filtering enabled with alpha = 0.4, and get -98 dBm on the input, while the last output value was -60 dBm. The new output would be: Avg[n] = 0.4 * Pwr[n] + 0.6 * Avg[n - 1] Avg[n] = (0.4 * -98) + (0.6 * -60) Avg[n] = -75.2 => around -75 Of course, this is not a silver bullet, but better than nothing. Change-Id: Ib6dcadbf14ef59696c6a546bd323bda92d399f17 Related: SYS#4916
2020-10-13 14:40:24 +00:00
/* We don't have 'Avg[n - 1]' if this is the first run */
if (*Avg100 == 0) {
power_control: implement EWMA based Uplink power filtering So far the Uplink power control loop did not filter the Uplink RSSI measurements (reported by the BTS) at all. The lack of filtering makes our implementation too quick on the trigger, so in the real deployments there will be unneeded Tx power oscillations. In order to reduce this effect, let's implement a very simple EWMA (also known as Single Pole IIR) filtering that is defined as follows: Avg[n] = a * Pwr[n] + (1 - a) * Avg[n - 1] where parameter 'a' determines how much weight of the latest UL RSSI measurement result 'Pwr[n]' carries vs the weight of the average 'Avg[n - 1]'. The value of 'a' is usually a float in range 0 .. 1, so: - value 0.5 gives equal weight to both 'Pwr[n]' and 'Avg[n - 1]'; - value 1.0 means no filtering at all (pass through); - value 0.0 makes no sense. This formula was further optimized with the use of '+=' operator. The floating point math was also eliminated by scaling everything up (by 100). For more details, see: https://en.wikipedia.org/wiki/Moving_average https://en.wikipedia.org/wiki/Low-pass_filter#Simple_infinite_impulse_response_filter https://tomroelandts.com/articles/low-pass-single-pole-iir-filter The EWMA filtering is now *enabled by default*, but can be disabled or (re-)configured over the VTY at any time: ! Completely disable filtering no uplink-power-filtering ! Enable EWMA smoothing with the given parameters uplink-power-filtering algo ewma beta <1-99> Note that the VTY command expects 'beta' instead of 'alpha': alpha = (100 - beta) and the value must be in %. This is done for simplicity: 1% means lowest smoothing, 99% means highest smoothing. Let's say we have EWMA filtering enabled with alpha = 0.4, and get -98 dBm on the input, while the last output value was -60 dBm. The new output would be: Avg[n] = 0.4 * Pwr[n] + 0.6 * Avg[n - 1] Avg[n] = (0.4 * -98) + (0.6 * -60) Avg[n] = -75.2 => around -75 Of course, this is not a silver bullet, but better than nothing. Change-Id: Ib6dcadbf14ef59696c6a546bd323bda92d399f17 Related: SYS#4916
2020-10-13 14:40:24 +00:00
*Avg100 = Pwr * EWMA_SCALE_FACTOR;
return Pwr;
}
*Avg100 += A * (Pwr - *Avg100 / EWMA_SCALE_FACTOR);
return *Avg100 / EWMA_SCALE_FACTOR;
}
/* Calculate a 'delta' value (for the given MS/BS power control state and parameters)
* to be applied to the current Tx power level to approach the target level. */
static int calc_delta(const struct bts_power_ctrl_params *params,
struct lchan_power_ctrl_state *state,
const int rxlev_dbm)
{
int rxlev_dbm_avg;
int delta;
/* Filter input value(s) to reduce unnecessary Tx power oscillations */
switch (params->pf_algo) {
case BTS_PF_ALGO_EWMA:
rxlev_dbm_avg = do_pf_ewma(params, state, rxlev_dbm);
break;
case BTS_PF_ALGO_NONE:
default:
/* No filtering (pass through) */
rxlev_dbm_avg = rxlev_dbm;
}
/* How many dBs measured power should be increased (+) or decreased (-)
* to reach expected power. */
delta = params->target_dbm - rxlev_dbm_avg;
/* Tolerate small deviations from 'rx-target' */
if (abs(delta) <= params->hysteresis_db)
return 0;
/* Don't ever change more than PWR_{LOWER,RAISE}_MAX_DBM during one loop
* iteration, i.e. reduce the speed at which the MS transmit power can
* change. A higher value means a lower level (and vice versa) */
if (delta > params->raise_step_max_db)
delta = params->raise_step_max_db;
else if (delta < -params->lower_step_max_db)
delta = -params->lower_step_max_db;
return delta;
}
/*! compute the new MS POWER LEVEL communicated to the MS and store it in lchan.
* \param lchan logical channel for which to compute (and in which to store) new power value.
* \param[in] ms_power_lvl MS Power Level received from Uplink L1 SACCH Header in SACCH block.
* \param[in] ul_rssi_dbm Signal level of the received SACCH block, in dBm.
*/
int lchan_ms_pwr_ctrl(struct gsm_lchan *lchan,
const uint8_t ms_power_lvl,
const int8_t ul_rssi_dbm)
{
struct gsm_bts_trx *trx = lchan->ts->trx;
struct gsm_bts *bts = trx->bts;
enum gsm_band band = bts->band;
int8_t new_power_lvl; /* TS 05.05 power level */
int8_t ms_dbm, new_dbm, current_dbm, bsc_max_dbm;
const struct bts_power_ctrl_params *params = &bts->ul_power_ctrl;
struct lchan_power_ctrl_state *state = &lchan->ms_power_ctrl;
if (!trx_ms_pwr_ctrl_is_osmo(trx))
return 0;
if (state->fixed)
return 0;
ms_dbm = ms_pwr_dbm(band, ms_power_lvl);
if (ms_dbm < 0) {
LOGPLCHAN(lchan, DLOOP, LOGL_NOTICE,
"Failed to calculate dBm for power ctl level %" PRIu8 " on band %s\n",
ms_power_lvl, gsm_band_name(band));
return 0;
}
bsc_max_dbm = ms_pwr_dbm(band, state->max);
if (bsc_max_dbm < 0) {
LOGPLCHAN(lchan, DLOOP, LOGL_NOTICE,
"Failed to calculate dBm for power ctl level %" PRIu8 " on band %s\n",
state->max, gsm_band_name(band));
return 0;
}
/* Calculate the new Tx power value (in dBm) */
new_dbm = ms_dbm + calc_delta(params, state, ul_rssi_dbm);
/* Make sure new_dbm is never negative. ms_pwr_ctl_lvl() can later on
cope with any unsigned dbm value, regardless of band minimal value. */
if (new_dbm < 0)
new_dbm = 0;
/* Don't ask for smaller ms power level than the one set by BSC upon RSL CHAN ACT */
if (new_dbm > bsc_max_dbm)
new_dbm = bsc_max_dbm;
new_power_lvl = ms_pwr_ctl_lvl(band, new_dbm);
if (new_power_lvl < 0) {
LOGPLCHAN(lchan, DLOOP, LOGL_NOTICE,
"Failed to retrieve power level for %" PRId8 " dBm on band %d\n",
new_dbm, band);
return 0;
}
if (state->current == new_power_lvl) {
LOGPLCHAN(lchan, DLOOP, LOGL_INFO, "Keeping MS power at control level %d, %d dBm "
"(rx-ms-pwr-lvl %" PRIu8 ", max-ms-pwr-lvl %" PRIu8 ", rx-current %d dBm, rx-target %d dBm)\n",
new_power_lvl, new_dbm, ms_power_lvl, state->max,
ul_rssi_dbm, params->target_dbm);
return 0;
}
current_dbm = ms_pwr_dbm(band, state->current);
LOGPLCHAN(lchan, DLOOP, LOGL_INFO, "%s MS power from control level %d (%d dBm) to %d, %d dBm "
"(rx-ms-pwr-lvl %" PRIu8 ", max-ms-pwr-lvl %" PRIu8 ", rx-current %d dBm, rx-target %d dBm)\n",
(new_dbm > current_dbm) ? "Raising" : "Lowering",
state->current, current_dbm, new_power_lvl, new_dbm,
ms_power_lvl, state->max, ul_rssi_dbm, params->target_dbm);
/* store the resulting new MS power level in the lchan */
state->current = new_power_lvl;
bts_model_adjst_ms_pwr(lchan);
return 1;
}
/*! compute the new Downlink attenuation value for the given logical channel.
* \param lchan logical channel for which to compute (and in which to store) new power value.
* \param[in] gh pointer to the beginning of (presumably) a Measurement Report.
*/
int lchan_bs_pwr_ctrl(struct gsm_lchan *lchan,
const struct gsm48_hdr *gh)
{
struct gsm_bts_trx *trx = lchan->ts->trx;
struct gsm_bts *bts = trx->bts;
uint8_t rxqual_full, rxqual_sub;
uint8_t rxlev_full, rxlev_sub;
uint8_t rxqual, rxlev;
int delta, new;
const struct bts_power_ctrl_params *params = &bts->dl_power_ctrl;
struct lchan_power_ctrl_state *state = &lchan->bs_power_ctrl;
/* Check if BS Power Control is enabled */
if (state->fixed)
return 0;
/* Check if this is a Measurement Report */
if (gh->proto_discr != GSM48_PDISC_RR)
return 0;
if (gh->msg_type != GSM48_MT_RR_MEAS_REP)
return 0;
/* Check if the measurement results are valid */
if ((gh->data[1] & 0x40) == 0x40) {
LOGPLCHAN(lchan, DLOOP, LOGL_DEBUG,
"The measurement results are not valid\n");
return 0;
}
/* See 3GPP TS 44.018, section 10.5.2.20 */
rxqual_full = (gh->data[2] >> 4) & 0x7;
rxqual_sub = (gh->data[2] >> 1) & 0x7;
rxlev_full = gh->data[0] & 0x3f;
rxlev_sub = gh->data[1] & 0x3f;
LOGPLCHAN(lchan, DLOOP, LOGL_DEBUG, "Rx DL Measurement Report: "
"RXLEV-FULL(%02u), RXQUAL-FULL(%u), "
"RXLEV-SUB(%02u), RXQUAL-SUB(%u), "
"DTx is %s => using %s\n",
rxlev_full, rxqual_full, rxlev_sub, rxqual_sub,
lchan->tch.dtx.dl_active ? "enabled" : "disabled",
lchan->tch.dtx.dl_active ? "SUB" : "FULL");
/* If DTx is active on Downlink, use the '-SUB' */
if (lchan->tch.dtx.dl_active) {
rxqual = rxqual_sub;
rxlev = rxlev_sub;
} else { /* ... otherwise use the '-FULL' */
rxqual = rxqual_full;
rxlev = rxlev_full;
}
/* Bit Error Rate > 0 => reduce by 2 */
if (rxqual > 0) {
LOGPLCHAN(lchan, DLOOP, LOGL_INFO, "Reducing Downlink attenuation "
"by half: %u -> %u dB due to RXQUAL %u > 0\n",
state->current, state->current / 2, rxqual);
state->current /= 2;
return 1;
}
/* Calculate a 'delta' for the current attenuation level */
delta = calc_delta(params, state, rxlev2dbm(rxlev));
/* Basic signal transmission / reception formula:
*
* RxLev = TxPwr - (PathLoss + TxAtt)
*
* Here we want to change RxLev at the MS side, so:
*
* RxLev + Delta = TxPwr - (PathLoss + TxAtt) + Delta
*
* The only parameter we can change here is TxAtt, so:
*
* RxLev + Delta = TxPwr - PathLoss - TxAtt + Delta
* RxLev + Delta = TxPwr - PathLoss - (TxAtt - Delta)
*/
new = state->current - delta;
if (new > state->max)
new = state->max;
if (new < 0)
new = 0;
if (state->current != new) {
LOGPLCHAN(lchan, DLOOP, LOGL_INFO, "Changing Downlink attenuation: "
"%u -> %u dB (maximum %u dB, target %d dBm, delta %d dB)\n",
state->current, new, state->max, params->target_dbm, delta);
state->current = new;
return 1;
} else {
LOGPLCHAN(lchan, DLOOP, LOGL_INFO, "Keeping Downlink attenuation "
"at %u dB (maximum %u dB, target %d dBm, delta %d dB)\n",
state->current, state->max, params->target_dbm, delta);
return 0;
}
}