Files
firmware/src/gps/GPSUpdateScheduling.cpp

119 lines
3.8 KiB
C++
Raw Normal View History

GPS Power State tidy-up (#4161) * Refactor GPSPowerState enum Identifies a case where the GPS hardware is awake, but an update is not yet desired * Change terminology * Clear old lock-time prediction on triple press * Use exponential smoothing to predict lock time * Rename averageLockTime to predictedLockTime * Attempt: Send PMREQ with duration 0 on MCU deep-sleep * Attempt 2: Send PMREQ with duration 0 on MCU deep-sleep * Revert "Attempt 2: Send PMREQ with duration 0 on MCU deep-sleep" This reverts commit 8b697cd2a445355dcfab5b33e0ce7a3128cab151. * Revert "Attempt: Send PMREQ with duration 0 on MCU deep-sleep" This reverts commit 9d29ec7603a88056b9115796b29b5023165a93bb. * Remove unused notifyGPSSleep Observable Handled with notifyDeepSleep, and enable() / disable() * WIP: simplify GPS power management An initial attempt only. * Honor #3e9e0fd * No-op when moving between GPS_IDLE and GPS_ACTIVE * Ensure U-blox GPS is awake to receive indefinite sleep command * Longer pause when waking U-blox to send sleep command * Actually implement soft and hard sleep.. * Dynamically estimate the threshold for GPS_HARDSLEEP * Fallback to GPS_HARDSLEEP, if GPS_SOFTSLEEP unsupported * Move "excessive search time" behavior to scheduler class * Minor logging adjustments * Promote log to warning * Gratuitous buffer clearing on boot * Fix inverted standby pin logic Specifically the standby pin for L76B, L76K and clones Discovered during T-Echo testing: totally broken function, probe method failing. * Remove redundant pin init Now handled by setPowerState * Replace max() with if statements Avoid those platform specific implementations.. * Trunk formatting New round of settings.json changes keep catching me out, have to remember to re-enable my "clang-format" for windows workaround. * Remove some asserts from setPowerState Original aim was to prevent sending a 0 second PMREQ to U-blox hardware as part of a timed sleep (GPS_HARDSLEEP, GPS_SOFTSLEEP). I'm not sure this is super important, and it feels tidier to just allow the 0 second sleeptime here, rather than fudge the sleeptime further up. * Fix an error determining whether GPS_SOFTSLEEP is supported * Clarify a log entry * Set PIN_STANDBY for MCU deep-sleep Required to reach TTGO's advertised 0.25mA sleep current for T-Echo. Without this change: ~6mA.
2024-07-11 15:26:43 +12:00
#include "GPSUpdateScheduling.h"
#include "Default.h"
// Mark the time when searching for GPS position begins
void GPSUpdateScheduling::informSearching()
{
searchStartedMs = millis();
}
// Mark the time when searching for GPS is complete,
// then update the predicted lock-time
void GPSUpdateScheduling::informGotLock()
{
searchEndedMs = millis();
LOG_DEBUG("Took %us to get lock", (searchEndedMs - searchStartedMs) / 1000);
GPS Power State tidy-up (#4161) * Refactor GPSPowerState enum Identifies a case where the GPS hardware is awake, but an update is not yet desired * Change terminology * Clear old lock-time prediction on triple press * Use exponential smoothing to predict lock time * Rename averageLockTime to predictedLockTime * Attempt: Send PMREQ with duration 0 on MCU deep-sleep * Attempt 2: Send PMREQ with duration 0 on MCU deep-sleep * Revert "Attempt 2: Send PMREQ with duration 0 on MCU deep-sleep" This reverts commit 8b697cd2a445355dcfab5b33e0ce7a3128cab151. * Revert "Attempt: Send PMREQ with duration 0 on MCU deep-sleep" This reverts commit 9d29ec7603a88056b9115796b29b5023165a93bb. * Remove unused notifyGPSSleep Observable Handled with notifyDeepSleep, and enable() / disable() * WIP: simplify GPS power management An initial attempt only. * Honor #3e9e0fd * No-op when moving between GPS_IDLE and GPS_ACTIVE * Ensure U-blox GPS is awake to receive indefinite sleep command * Longer pause when waking U-blox to send sleep command * Actually implement soft and hard sleep.. * Dynamically estimate the threshold for GPS_HARDSLEEP * Fallback to GPS_HARDSLEEP, if GPS_SOFTSLEEP unsupported * Move "excessive search time" behavior to scheduler class * Minor logging adjustments * Promote log to warning * Gratuitous buffer clearing on boot * Fix inverted standby pin logic Specifically the standby pin for L76B, L76K and clones Discovered during T-Echo testing: totally broken function, probe method failing. * Remove redundant pin init Now handled by setPowerState * Replace max() with if statements Avoid those platform specific implementations.. * Trunk formatting New round of settings.json changes keep catching me out, have to remember to re-enable my "clang-format" for windows workaround. * Remove some asserts from setPowerState Original aim was to prevent sending a 0 second PMREQ to U-blox hardware as part of a timed sleep (GPS_HARDSLEEP, GPS_SOFTSLEEP). I'm not sure this is super important, and it feels tidier to just allow the 0 second sleeptime here, rather than fudge the sleeptime further up. * Fix an error determining whether GPS_SOFTSLEEP is supported * Clarify a log entry * Set PIN_STANDBY for MCU deep-sleep Required to reach TTGO's advertised 0.25mA sleep current for T-Echo. Without this change: ~6mA.
2024-07-11 15:26:43 +12:00
updateLockTimePrediction();
}
// Clear old lock-time prediction data.
// When re-enabling GPS with user button.
void GPSUpdateScheduling::reset()
{
searchStartedMs = 0;
searchEndedMs = 0;
searchCount = 0;
predictedMsToGetLock = 0;
}
// How many milliseconds before we should next search for GPS position
// Used by GPS hardware directly, to enter timed hardware sleep
uint32_t GPSUpdateScheduling::msUntilNextSearch()
{
uint32_t now = millis();
// Target interval (seconds), between GPS updates
uint32_t updateInterval = Default::getConfiguredOrDefaultMs(config.position.gps_update_interval, default_gps_update_interval);
// Check how long until we should start searching, to hopefully hit our target interval
uint32_t dueAtMs = searchEndedMs + updateInterval;
uint32_t compensatedStart = dueAtMs - predictedMsToGetLock;
int32_t remainingMs = compensatedStart - now;
// If we should have already started (negative value), start ASAP
if (remainingMs < 0)
remainingMs = 0;
return (uint32_t)remainingMs;
}
// How long have we already been searching?
// Used to abort a search in progress, if it runs unacceptably long
GPS Power State tidy-up (#4161) * Refactor GPSPowerState enum Identifies a case where the GPS hardware is awake, but an update is not yet desired * Change terminology * Clear old lock-time prediction on triple press * Use exponential smoothing to predict lock time * Rename averageLockTime to predictedLockTime * Attempt: Send PMREQ with duration 0 on MCU deep-sleep * Attempt 2: Send PMREQ with duration 0 on MCU deep-sleep * Revert "Attempt 2: Send PMREQ with duration 0 on MCU deep-sleep" This reverts commit 8b697cd2a445355dcfab5b33e0ce7a3128cab151. * Revert "Attempt: Send PMREQ with duration 0 on MCU deep-sleep" This reverts commit 9d29ec7603a88056b9115796b29b5023165a93bb. * Remove unused notifyGPSSleep Observable Handled with notifyDeepSleep, and enable() / disable() * WIP: simplify GPS power management An initial attempt only. * Honor #3e9e0fd * No-op when moving between GPS_IDLE and GPS_ACTIVE * Ensure U-blox GPS is awake to receive indefinite sleep command * Longer pause when waking U-blox to send sleep command * Actually implement soft and hard sleep.. * Dynamically estimate the threshold for GPS_HARDSLEEP * Fallback to GPS_HARDSLEEP, if GPS_SOFTSLEEP unsupported * Move "excessive search time" behavior to scheduler class * Minor logging adjustments * Promote log to warning * Gratuitous buffer clearing on boot * Fix inverted standby pin logic Specifically the standby pin for L76B, L76K and clones Discovered during T-Echo testing: totally broken function, probe method failing. * Remove redundant pin init Now handled by setPowerState * Replace max() with if statements Avoid those platform specific implementations.. * Trunk formatting New round of settings.json changes keep catching me out, have to remember to re-enable my "clang-format" for windows workaround. * Remove some asserts from setPowerState Original aim was to prevent sending a 0 second PMREQ to U-blox hardware as part of a timed sleep (GPS_HARDSLEEP, GPS_SOFTSLEEP). I'm not sure this is super important, and it feels tidier to just allow the 0 second sleeptime here, rather than fudge the sleeptime further up. * Fix an error determining whether GPS_SOFTSLEEP is supported * Clarify a log entry * Set PIN_STANDBY for MCU deep-sleep Required to reach TTGO's advertised 0.25mA sleep current for T-Echo. Without this change: ~6mA.
2024-07-11 15:26:43 +12:00
uint32_t GPSUpdateScheduling::elapsedSearchMs()
{
// If searching
if (searchStartedMs > searchEndedMs)
return millis() - searchStartedMs;
// If not searching - 0ms. We shouldn't really consume this value
else
return 0;
}
// Is it now time to begin searching for a GPS position?
bool GPSUpdateScheduling::isUpdateDue()
{
return (msUntilNextSearch() == 0);
}
// Have we been searching for a GPS position for too long?
bool GPSUpdateScheduling::searchedTooLong()
{
2024-10-28 06:40:48 -05:00
uint32_t minimumOrConfiguredSecs =
Default::getConfiguredOrMinimumValue(config.position.position_broadcast_secs, default_broadcast_interval_secs);
uint32_t maxSearchMs = Default::getConfiguredOrDefaultMs(minimumOrConfiguredSecs, default_broadcast_interval_secs);
GPS Power State tidy-up (#4161) * Refactor GPSPowerState enum Identifies a case where the GPS hardware is awake, but an update is not yet desired * Change terminology * Clear old lock-time prediction on triple press * Use exponential smoothing to predict lock time * Rename averageLockTime to predictedLockTime * Attempt: Send PMREQ with duration 0 on MCU deep-sleep * Attempt 2: Send PMREQ with duration 0 on MCU deep-sleep * Revert "Attempt 2: Send PMREQ with duration 0 on MCU deep-sleep" This reverts commit 8b697cd2a445355dcfab5b33e0ce7a3128cab151. * Revert "Attempt: Send PMREQ with duration 0 on MCU deep-sleep" This reverts commit 9d29ec7603a88056b9115796b29b5023165a93bb. * Remove unused notifyGPSSleep Observable Handled with notifyDeepSleep, and enable() / disable() * WIP: simplify GPS power management An initial attempt only. * Honor #3e9e0fd * No-op when moving between GPS_IDLE and GPS_ACTIVE * Ensure U-blox GPS is awake to receive indefinite sleep command * Longer pause when waking U-blox to send sleep command * Actually implement soft and hard sleep.. * Dynamically estimate the threshold for GPS_HARDSLEEP * Fallback to GPS_HARDSLEEP, if GPS_SOFTSLEEP unsupported * Move "excessive search time" behavior to scheduler class * Minor logging adjustments * Promote log to warning * Gratuitous buffer clearing on boot * Fix inverted standby pin logic Specifically the standby pin for L76B, L76K and clones Discovered during T-Echo testing: totally broken function, probe method failing. * Remove redundant pin init Now handled by setPowerState * Replace max() with if statements Avoid those platform specific implementations.. * Trunk formatting New round of settings.json changes keep catching me out, have to remember to re-enable my "clang-format" for windows workaround. * Remove some asserts from setPowerState Original aim was to prevent sending a 0 second PMREQ to U-blox hardware as part of a timed sleep (GPS_HARDSLEEP, GPS_SOFTSLEEP). I'm not sure this is super important, and it feels tidier to just allow the 0 second sleeptime here, rather than fudge the sleeptime further up. * Fix an error determining whether GPS_SOFTSLEEP is supported * Clarify a log entry * Set PIN_STANDBY for MCU deep-sleep Required to reach TTGO's advertised 0.25mA sleep current for T-Echo. Without this change: ~6mA.
2024-07-11 15:26:43 +12:00
// If broadcast interval set to max, no such thing as "too long"
if (maxSearchMs == UINT32_MAX)
return false;
// If we've been searching longer than our position broadcast interval: that's too long
else if (elapsedSearchMs() > maxSearchMs)
return true;
// Otherwise, not too long yet!
else
return false;
}
// Updates the predicted time-to-get-lock, by exponentially smoothing the latest observation
void GPSUpdateScheduling::updateLockTimePrediction()
{
// How long did it take to get GPS lock this time?
// Duration between down() calls
int32_t lockTime = searchEndedMs - searchStartedMs;
if (lockTime < 0)
lockTime = 0;
// Ignore the first lock-time: likely to be long, will skew data
// Second locktime: likely stable. Use to initialize the smoothing filter
GPS Power State tidy-up (#4161) * Refactor GPSPowerState enum Identifies a case where the GPS hardware is awake, but an update is not yet desired * Change terminology * Clear old lock-time prediction on triple press * Use exponential smoothing to predict lock time * Rename averageLockTime to predictedLockTime * Attempt: Send PMREQ with duration 0 on MCU deep-sleep * Attempt 2: Send PMREQ with duration 0 on MCU deep-sleep * Revert "Attempt 2: Send PMREQ with duration 0 on MCU deep-sleep" This reverts commit 8b697cd2a445355dcfab5b33e0ce7a3128cab151. * Revert "Attempt: Send PMREQ with duration 0 on MCU deep-sleep" This reverts commit 9d29ec7603a88056b9115796b29b5023165a93bb. * Remove unused notifyGPSSleep Observable Handled with notifyDeepSleep, and enable() / disable() * WIP: simplify GPS power management An initial attempt only. * Honor #3e9e0fd * No-op when moving between GPS_IDLE and GPS_ACTIVE * Ensure U-blox GPS is awake to receive indefinite sleep command * Longer pause when waking U-blox to send sleep command * Actually implement soft and hard sleep.. * Dynamically estimate the threshold for GPS_HARDSLEEP * Fallback to GPS_HARDSLEEP, if GPS_SOFTSLEEP unsupported * Move "excessive search time" behavior to scheduler class * Minor logging adjustments * Promote log to warning * Gratuitous buffer clearing on boot * Fix inverted standby pin logic Specifically the standby pin for L76B, L76K and clones Discovered during T-Echo testing: totally broken function, probe method failing. * Remove redundant pin init Now handled by setPowerState * Replace max() with if statements Avoid those platform specific implementations.. * Trunk formatting New round of settings.json changes keep catching me out, have to remember to re-enable my "clang-format" for windows workaround. * Remove some asserts from setPowerState Original aim was to prevent sending a 0 second PMREQ to U-blox hardware as part of a timed sleep (GPS_HARDSLEEP, GPS_SOFTSLEEP). I'm not sure this is super important, and it feels tidier to just allow the 0 second sleeptime here, rather than fudge the sleeptime further up. * Fix an error determining whether GPS_SOFTSLEEP is supported * Clarify a log entry * Set PIN_STANDBY for MCU deep-sleep Required to reach TTGO's advertised 0.25mA sleep current for T-Echo. Without this change: ~6mA.
2024-07-11 15:26:43 +12:00
if (searchCount == 1)
predictedMsToGetLock = lockTime;
// Third locktime and after: predict using exponential smoothing. Respond slowly to changes
else if (searchCount > 1)
predictedMsToGetLock = (lockTime * weighting) + (predictedMsToGetLock * (1 - weighting));
searchCount++; // Only tracked so we can disregard initial lock-times
GPS Power State tidy-up (#4161) * Refactor GPSPowerState enum Identifies a case where the GPS hardware is awake, but an update is not yet desired * Change terminology * Clear old lock-time prediction on triple press * Use exponential smoothing to predict lock time * Rename averageLockTime to predictedLockTime * Attempt: Send PMREQ with duration 0 on MCU deep-sleep * Attempt 2: Send PMREQ with duration 0 on MCU deep-sleep * Revert "Attempt 2: Send PMREQ with duration 0 on MCU deep-sleep" This reverts commit 8b697cd2a445355dcfab5b33e0ce7a3128cab151. * Revert "Attempt: Send PMREQ with duration 0 on MCU deep-sleep" This reverts commit 9d29ec7603a88056b9115796b29b5023165a93bb. * Remove unused notifyGPSSleep Observable Handled with notifyDeepSleep, and enable() / disable() * WIP: simplify GPS power management An initial attempt only. * Honor #3e9e0fd * No-op when moving between GPS_IDLE and GPS_ACTIVE * Ensure U-blox GPS is awake to receive indefinite sleep command * Longer pause when waking U-blox to send sleep command * Actually implement soft and hard sleep.. * Dynamically estimate the threshold for GPS_HARDSLEEP * Fallback to GPS_HARDSLEEP, if GPS_SOFTSLEEP unsupported * Move "excessive search time" behavior to scheduler class * Minor logging adjustments * Promote log to warning * Gratuitous buffer clearing on boot * Fix inverted standby pin logic Specifically the standby pin for L76B, L76K and clones Discovered during T-Echo testing: totally broken function, probe method failing. * Remove redundant pin init Now handled by setPowerState * Replace max() with if statements Avoid those platform specific implementations.. * Trunk formatting New round of settings.json changes keep catching me out, have to remember to re-enable my "clang-format" for windows workaround. * Remove some asserts from setPowerState Original aim was to prevent sending a 0 second PMREQ to U-blox hardware as part of a timed sleep (GPS_HARDSLEEP, GPS_SOFTSLEEP). I'm not sure this is super important, and it feels tidier to just allow the 0 second sleeptime here, rather than fudge the sleeptime further up. * Fix an error determining whether GPS_SOFTSLEEP is supported * Clarify a log entry * Set PIN_STANDBY for MCU deep-sleep Required to reach TTGO's advertised 0.25mA sleep current for T-Echo. Without this change: ~6mA.
2024-07-11 15:26:43 +12:00
LOG_DEBUG("Predict %us to get next lock", predictedMsToGetLock / 1000);
GPS Power State tidy-up (#4161) * Refactor GPSPowerState enum Identifies a case where the GPS hardware is awake, but an update is not yet desired * Change terminology * Clear old lock-time prediction on triple press * Use exponential smoothing to predict lock time * Rename averageLockTime to predictedLockTime * Attempt: Send PMREQ with duration 0 on MCU deep-sleep * Attempt 2: Send PMREQ with duration 0 on MCU deep-sleep * Revert "Attempt 2: Send PMREQ with duration 0 on MCU deep-sleep" This reverts commit 8b697cd2a445355dcfab5b33e0ce7a3128cab151. * Revert "Attempt: Send PMREQ with duration 0 on MCU deep-sleep" This reverts commit 9d29ec7603a88056b9115796b29b5023165a93bb. * Remove unused notifyGPSSleep Observable Handled with notifyDeepSleep, and enable() / disable() * WIP: simplify GPS power management An initial attempt only. * Honor #3e9e0fd * No-op when moving between GPS_IDLE and GPS_ACTIVE * Ensure U-blox GPS is awake to receive indefinite sleep command * Longer pause when waking U-blox to send sleep command * Actually implement soft and hard sleep.. * Dynamically estimate the threshold for GPS_HARDSLEEP * Fallback to GPS_HARDSLEEP, if GPS_SOFTSLEEP unsupported * Move "excessive search time" behavior to scheduler class * Minor logging adjustments * Promote log to warning * Gratuitous buffer clearing on boot * Fix inverted standby pin logic Specifically the standby pin for L76B, L76K and clones Discovered during T-Echo testing: totally broken function, probe method failing. * Remove redundant pin init Now handled by setPowerState * Replace max() with if statements Avoid those platform specific implementations.. * Trunk formatting New round of settings.json changes keep catching me out, have to remember to re-enable my "clang-format" for windows workaround. * Remove some asserts from setPowerState Original aim was to prevent sending a 0 second PMREQ to U-blox hardware as part of a timed sleep (GPS_HARDSLEEP, GPS_SOFTSLEEP). I'm not sure this is super important, and it feels tidier to just allow the 0 second sleeptime here, rather than fudge the sleeptime further up. * Fix an error determining whether GPS_SOFTSLEEP is supported * Clarify a log entry * Set PIN_STANDBY for MCU deep-sleep Required to reach TTGO's advertised 0.25mA sleep current for T-Echo. Without this change: ~6mA.
2024-07-11 15:26:43 +12:00
}
// How long do we expect to spend searching for a lock?
uint32_t GPSUpdateScheduling::predictedSearchDurationMs()
{
return GPSUpdateScheduling::predictedMsToGetLock;
}