September 1998, Issue 98

PIC'Spectrum
Audio Spectrum Analyzer

DEVELOPMENT AND DEBUG

Debugging a signal-processing embedded application isn’t easy. And, I don’t have access to an adequate ICE. Since it might be useful for similar projects, let me briefly explain my methodology (which must not be too bad: PIC’Spectrum actually works!).

After some timing calculations to ensure the feasibility of the concept, I started with a prototyping phase on a PC using a PC sound board as input. I used floating-point calculations, with everything in C (see Listing 1). I then translated the software to use exclusively integer numbers and debugged it.

The third step was to develop a fixed-point library (still on the PC) that prefigures the future PIC-based fixed-point library. I also modified the FFT code to do all calculations exclusively through this library. It was quite easy to write this fixed-point library in PIC assembler, debug it on Microchip’s simulator, and translate the FFT routine from C to assembler.

Using the simulator is useful for numerical-calculation software. In fact, it was possible to open two windows on my PC, the first being a standard PC debugger with my C-code FFT, and the second the PIC simulator with the manually translated assembler code. Single stepping between the two codes with a test signal as an input helped me quickly find the more tricky bugs.

When the signal-processing part was completed and debugged, I wrote the real-time part (signal acquisition and video generation) and debugged it as much as I could on the simulator.

This phase was helpful for correcting timing issues related to video synchronization. For example, with a breakpoint set on each access to the horizontal synchronization pin, it’s easy to verify if the timings on this pin meet the VGA specification.

I waited until this phase was successful before cabling the prototype. The first EPROM I burned didn’t work 100%, but I got a working video image and something resembling a spectrum display.

Thanks to the in-circuit programming and some debug pins, the software was working soon. In fact, 90% of the bugs found in this last step were related to banking register issues, which are tough to simulate because hardware ports are involved.