Basic AC wattage question: am I doing my math right?

Please elaborate on that. I can understand that a multiplying A/D might be linear enough at low levels to produce an output that could be integrated over time to give a better accuracy than 1 LSB, but a system that uses digital processing for both current and voltage needs more bits to produce a meaningful output. This is even more so for complex waveforms, or where the voltage is AC and the current is DC. But the main practical limiting factors at low levels are nonlinearity, phase shift, distortion, and offset, especially in the current measuring components.

Paul

Sample the voltage and current waveforms with, say, an 8 or 10-bit ADC, preferably simultaneously, but at least not many microseconds apart. The voltage waveform is usually pretty constant in amplitude in regular AC systems, so adc resolution is not a problem there. The current waveform can have a huge dynamic range, 10,000:1 or so for a metering-quality measurement. So add a few lsb's of noise to the current signal before digitizing it; that smears out the quantization errors. Now software autozero the current samples to take out any dc offsets, to a fraction of an lsb, say, 16 bits or so. Now multiply the zeroed current sample with the paired voltage sample and average the product points to get power, integrate to get energy. The statistics are great if you do the math properly. A few watts resolution out of, say, 20 kw full-scale is possible.

Actually, DC offset in the current signal washes out when it's multiplied by the voltage sine wave samples and averaged to make power, as long as the voltage signals doesn't have offset too. Software autozering both the voltage and current data allow you to use unipolar, unsigned adc's and not wory about residual (or huge) dc offsets in the signal conditioning or the adc itself.

The worst low-power error will be crosstalk between the voltage and current signals, which can happen magnetically, or in the adc mux, or any number of other interesting places. Any crosstalk or correlated noise does produce a power offset, some of which can be software fudged out.

John