Case Study 3: Algorithm Design
DSP on the 8052
The commercial viability of a client's product was prejudiced by its susceptibility to vibration.
Attempts to filter the input using analogue techniques had been unsuccesful.
ADAXO's proposal was to apply digital processing to the output data.
This suggestion was viewed with suspicion by the client who had been advised that digital
signal processing (DSP) would be too expensive and power hungry.
ADAXO's innovative approach was to implement a solution based on the HAAR transform.
Demonstration
A convincing demonstration on a PC using data captured from the original
design was the first step.
The single point data error is evidence that the original design was extremely
susceptible to electromagnetic interference. This issue was dealt with separately so it
is not considered further in this document.
Performance
The step response of the instrument shows some overshoot which is inherent in its design.
The algorithm identifies it reliably and ceases to flag valid data until the
input has settled to its new value.
The instrument was intended for use in a factory environment but the next figure
shows the effect of dragging a rubber handled screwdriver lightly along
the work piece.
The algorithm responds promptly to the change in the statistics of the input data.
It continues to develop a smoothed output (actually an 8pt FIR
which comes "free" with the transform!) while the data are flagged as invalid.
The valid flag is set only when the noise is removed and the
FIR filter has been flushed.
Implementation
To suit the application and adapt to the 8052 the transform was modified:
- The "Fast HAAR transform" was configured as a cyclic process operating
on a circular buffer in IDATA space.
- Coefficients were constrained to +1, 0, -1.
eliminating the need for floating point multiplication.
Conclusion
In its original form sensitivity to the slightest mechanical shock made the instrument unusable.
Processing the data digitally made it commercially viable and achieved significant
savings in manufacturing cost.
