multiple tickers stop working

[andrewjfox]

Hi, I've read about previous issues with tickers being resolved, however I'm having trouble and want to confirm if ticker functions are all ok. I'm using nRF51422 based board and nRF51-DK platform on mbed to generate FOTA files.

Code has BLE running. Problem occurs after several seconds when connected, or after 10's seconds when not connected.

Following is class used to drive RGB leds with single ticker for each led (there are 2 leds

#include "mbed.h"

class RGBColour
{
public:
    RGBColour(int r=0, int g=0, int b=0) {
        red = r;
        green = g;
        blue = b;
        brightness = 100;
    }    

    bool isOn(void) {
        if(!brightness) return false;
        if(!red && !green && !blue) return false;
        return true;
    }    

    int red;        // 0 - 100 %
    int green;
    int blue;
    int brightness; // 0 - 100 %
};



class RGBControl {
public:
#ifdef TARGET_DK
    const static int LED_ON    = 0;
    const static int LED_OFF   = 1;
#else
    const static int LED_ON    = 1;
    const static int LED_OFF   = 0;
#endif

    // 20ms PWM period with ? count resolution on RGB values
    // any faster and interrupts can't service in time
    const static float PWM_PERIOD = 0.025;
    const static int COUNTS_PER_PERIOD = 10;
    const static float SEC_PER_COUNT = (PWM_PERIOD / (float)COUNTS_PER_PERIOD);

    RGBControl(PinName red, PinName green, PinName blue) : _red(red), _green(green), _blue(blue) {  
        rgb(RGBColour(0,0,0)); 
    }

    /** @param RGB values
    */        
    void rgb(RGBColour c) {        
        _ticker.detach();
        _red = LED_OFF;
        _green = LED_OFF;
        _blue = LED_OFF;

        if(c.isOn()) {
            _tickerCompare[0] = (((c.red * c.brightness) / 100) * COUNTS_PER_PERIOD) / 100;
            _tickerCompare[1] = (((c.green * c.brightness) / 100) * COUNTS_PER_PERIOD) / 100;
            _tickerCompare[2] = (((c.blue * c.brightness) / 100) * COUNTS_PER_PERIOD) / 100;

            _tickerCount = COUNTS_PER_PERIOD;
            _ticker.attach(this, &RGBControl::tickerCallback, SEC_PER_COUNT);
        }    
    }

private:
    DigitalOut _red;
    DigitalOut _green;
    DigitalOut _blue;
    Ticker _ticker;
    int _tickerCount;
    int _tickerCompare[3];

    void tickerCallback(void)
    {
        if(++_tickerCount >= COUNTS_PER_PERIOD) { 
            _tickerCount=0;

            if(_tickerCompare[0]) _red = LED_ON;
            if(_tickerCompare[1]) _green = LED_ON;
            if(_tickerCompare[2]) _blue = LED_ON;
        }    

        if(_tickerCount == _tickerCompare[0]) _red = LED_OFF;
        if(_tickerCount == _tickerCompare[1]) _green = LED_OFF;
        if(_tickerCount == _tickerCompare[2]) _blue = LED_OFF;
    }  
};

if I drive a single LED, all good, if I drive 2 leds simultaneously, both of the tickers stop after several seconds.

RGBControl led1(LED1_RED,LED1_GREEN,LED1_BLUE);
RGBControl led2(LED2_RED,LED2_GREEN,LED2_BLUE);
...
while(1)
{
...
if(triggerEvent) {
    // call this every second, increments through colour table
    led1.rgb(colour[currentColour]);
    if(++currentColour >= COLOUR_MAX) currentColour = COLOUR_BLACK;
    led2.rgb(colour[currentColour]);

    triggerEvent = false;
}
...
}

Is there an issue with my code, or is it possibly an unresolved ticker issue?

Andrew

Some additional background:

• my code has 1/2 doz other tickers and timers in main loop for various functions, but are all quite slow and haven't noticed any issues until led class started freezing
• BLE functions keep working in background even when led class freezes
• other gpio functions, that read inputs using InterruptIn and have timers for debouncing, also stop working when leds freeze

[anpilog]

Hi,

I guess you are running Ticker at too high frequency.
NRF51 platform use RTC for us_ticker and it's just 32kHz.
Besides there was old issue with us_tick_timer for NRF51 and some STM32 boards when us_ticker stuck under high load.
As I remember the issue on NRF51 target was resolved by using SDK API for RTC.

Anyway I would suggest to do high load test for us_ticker first.

Regards

[andrewjfox]

That's what I thought at first, but it's only running with about 2msec ticker interval, or 500Hz.

@andrewjfox : do the interrupts stop working altogether? or does it just fail with the timer part?

I'm having a similar problem with the RemoteIR library on two different NRF51822 boards, I see the interrupts still fire, just nothing happens.

[andrewjfox]

Not sure if all interrupts stop, since the BLE core is still partially functioning (it advertises and responds to connection request, just doesn't update characterisitic values from application code) however all of the interrupts attached to my application code stop. I have several inputs setup as InterruptIn and they stop responding. Don't know if the interrupts are firing, and code not running, or if interrupts stop firing.

Sometimes after about 20secs it all comes back to life, but most of the time I have to manually reset, which is a real pain since it's a BLE battery device and doesn't have an easy way of manually resetting, so I have to short the power pins.

oh, i was gonna ask if you could say put a simple led toggle on there, but I guess that'd be a little difficult.

[sander]

@andrewjfox, did you find a solution? I’m seeing similar delays and am also using multiple tickers and timeouts in my nRF51822 board.

[andrewjfox]

No, no solution as yet. I've had to disable the PWM function and just have simple on/off control for LEDs until I can resolve the problem. I'm seeing timing inconsistencies more often now as I implement more features in the code, although I'm being really careful to not add more timers unless needed and keep their periods long to minimise number of interrupts. The issues all started once I included the bluetooth functions. My suspicion is that the softdevice uses a lot of system resources so that there's very little left for the rest of the device.

If you find a solution let me know

@andrewjfox so you've narrowed the problem to pwm? I didn't realize you were using PWM, my eyes skipped over the period/pwm mentions because i didn't see usage of PwmOut and only DigitalOut

narrowed it is not right, since you're still having timing problems! I do wonder though if there are pwm specific issues in interaction with the softdevice though.

on that note, I'm not sure why the nordic sdk components in mbed don't include the nordic pwm api

[andrewjfox]

Apologies for the confusion. I meant the software PWM I've implemented using the timers, not the mbed PWM objects.

Yes I agree that it would be worth including the nordic pwm api. I've tried several times to include it as a library in my project, but can't get the project to compile.

so why didn't you use the (mbed) native pwmout interface?

[andrewjfox]

good question, 2 reasons,

1. Only 3 available and I'm using 1 for a buzzer output, and the 2 RGB LEDs are driven seperately so need 6 PWMs (2 x Red, Green and Blue). They're driven separately rather than multiplexed for other design reasons.
2. I'm trying to reduce power and discovered that the mbed Pwm objects can't be deleted (bug in api that doesn't delete deconstruct correctly). Even when not required the PWM timer is still running in background so uses power.
   2b I tried driving buzzer using timers, to be able to save power when not required, but as it turns out I have to use at least 1 Pwm object for the buzzer output because the timers can't keep up to drive at reasonable audio frequency around 1kHz, so I'm not saving power anyway. But that's a separate problem

I think they removed the 3 available restriction, but made them all share something (i can't recall which setting)

can you file a bug about the your number 2 issue?