Erik:
The "worthless" technigue may actually work as shown IF the algorithm is not used for more than 12 bits AND if the 12 bit samples are left adjusted in the 16 bit words.

But then why not add up just the 12 bit samples as right adjusted into a 16 bit sum and then simply divide that sum by shifting right 3 times. (You can fit 16 12 bit samples summed together into a 16 bit result without overflow).

Steve:
I do not have 8052 code for the running sum averaging technigue. I have used it on multiple platforms though. Once on a 68HC11 in assembler, in C code on a DOS PC platform, and on a TI TMS320C31 DSP platform. The algorithm is however easy to code in either C or assembler language. It is assembler in particular where having the buffer an even power of two in size is handy because even the index to the load point in the sample table can be wrapped by a simple AND mask operation.

Michael Karas