A frequency meter, when measuring in the 0->2KHz range (I.E. 0.000 -> 1.999 on a 3 1/2 digit device) will often sample for 10 seconds between display updates.

If you have a carefully designed setup such that you have an accurate 1 second sampling window then the error you should see in the capture process cannot ever be more accurate than exact or minus one count. If on the other hand your 1 second gating window itself is not accurate then the 1 second gate time could be subject to its own error. If the gate was generated from the same 1KHz frequency that you are trying to measure then that accounts for your exact to -3 reading range.

So, that said, you need to strive to find a way to make your gate time exactly 1 second. Then you need to find a way to couple the gate time to the frequency measuring capture process enable in a direct way as opposed to coupling through software in a polling or interrupt driven process.

When measuring frequency that is asynchronous to the gate time there is always the possibility of a concurrency between the gate edge and the clock edge to the capture counter. This may lead to a metastable consition on the capture counter. On most processors I have worked with this condition this can cause the counter to fail to increment during that one count time.

Finally note that frequency measuring devices rarely "average" the displays in the normal way that you expect. Instead the gate time is made 10 times (or more) longer than the resolution in frequency of the lowest order digit in the display. In other words the averaging is done in the accumulation process.

Michael Karas