Question: 32K FFT one application - 8-bit complex data. the other application - 16-bit complex data... what kind of can of error can you expect in both cases.

Answer: In radix 2 and radix 4 FFT's one bit per stage of resolution is lost. Radix 4 has 1/2 half the number of stages. If a mixed radix (16K radix 4 --- 7 stages ) followed by 1 radix 2 stage is used. The result should have lost 8 bits of precision. This means to achieve a 8-bit result you must start with 16-bit data and coefficients (or scale appropriately to 16-bits.)