Posted: Wed Jul 16, 2008 9:36 am
I should have said dv/dt , or delta voltage / delta time. Delta meaning in this case 'change of'. In this application, it's the ability of the optoisolator to give the correct output whilst the voltage difference between the input and output are changing. This is probably not a problem between cells, but between the pack as a whole and the master unit, can cause data loss or corruption.
Another concern is that of accuracy and resolution. 30mV resolution might be sufficient, but how accurate is that reading? If the units vary between each other by just plus or minus one 30mV step of resolution, that's nearly 0.1V which is starting to be a significant amount with these cells. The variation may well be more, although you can overcome that by calibrating each unit, storing a correction factor in memory. Generally a 10:1 ratio between accuracy and resolution is used. Don't assume you can select tighter tollerance components from a batch, the manufacturer may already have done so, leaving you with a dip in the middle of the standard distribution 'bell' curve.
As for filtering, you're going to need both analogue and digital filtering. A big problem with digital sampling systems is that of aliasing. Imagine you have a 100Hz square wave of 1V peak to peak. If you sampled it at 10mS intervals (100Hz) then depending on where in the waveform phase you sampled, you could get a reading of either 1V or 0V, whereas the average voltage is 500mV. With a passive filter, you want to have a cutoff frequency considerably less than the PWM frequency, but high enough not to give a sluggish response. With digital filtering, you want to take samples at the fastest convenient rate and then average them, which will help to remove the noise inherant to the A2D converter and reference voltage. You could simply take ten readings and then average them, or you might want a more sophisticated scheme, with a rolling average of the last ten samples and even a system that discards new samples that are too far different from the previous sample.
Another concern is that of accuracy and resolution. 30mV resolution might be sufficient, but how accurate is that reading? If the units vary between each other by just plus or minus one 30mV step of resolution, that's nearly 0.1V which is starting to be a significant amount with these cells. The variation may well be more, although you can overcome that by calibrating each unit, storing a correction factor in memory. Generally a 10:1 ratio between accuracy and resolution is used. Don't assume you can select tighter tollerance components from a batch, the manufacturer may already have done so, leaving you with a dip in the middle of the standard distribution 'bell' curve.
As for filtering, you're going to need both analogue and digital filtering. A big problem with digital sampling systems is that of aliasing. Imagine you have a 100Hz square wave of 1V peak to peak. If you sampled it at 10mS intervals (100Hz) then depending on where in the waveform phase you sampled, you could get a reading of either 1V or 0V, whereas the average voltage is 500mV. With a passive filter, you want to have a cutoff frequency considerably less than the PWM frequency, but high enough not to give a sluggish response. With digital filtering, you want to take samples at the fastest convenient rate and then average them, which will help to remove the noise inherant to the A2D converter and reference voltage. You could simply take ten readings and then average them, or you might want a more sophisticated scheme, with a rolling average of the last ten samples and even a system that discards new samples that are too far different from the previous sample.