Guitar Effects Pitch Scaling
|
DSP Implementation of SOLA As the samples arrive from the AIC, they are resampled using multirate signal processing. Upsampling and interpolation can occur as soon as two samples are present. The new upsampled part is written into memory after they have been through an antialiasing filter to prevent frequency folding from the downsampling process. When there are enough samples, downsampling takes place and the data is ready for grain processing. If the resampling process results in sample reduction (higher pitch), then downsampling cannot take place immediately after an upsample. It must wait until there are more upsampled points available than the downsampling factor. This leads to requiring a small buffer of length that depends on the difference between upsampling and downsampling factors. At +5 semitone scaling, this results in a delay of 42ms. If resampling is decreasing the pitch, then there are always enough samples to downsample immediately after an upsample. After the signal has been resampled, it is split up into four parts. Each of these parts is processed by the same code, but each as a separate grain. Therefore, four grains are continually being processed, each with its own index position in the grain. This allows for a 75% overlap of grains. This means that each sample is multiplied by four different gain factors for the triangular windowing process. These gain factors are incremented or decremented by 33 for every new sample that arrives. If the grain position is less than 2040 (half way) then the gain factor is incremented. Otherwise, it is decremented and this forms the triangular windowing function. Once a new grain has begun, and 100 samples have been processed, cross correlation is performed between the current and preceding grains. This requires that as a grain is passing over its 25% mark (where the next grain will be overlapped and added), it records 100 samples for this purpose. The best alignment for the grains is used to instruct where the current grain starts. This whole process adds a 4ms delay to the system. The processed data from the grains is constantly being added together and scaled to achieve an output data stream that is sent to the AIC. Figure 25 shows the implementation method in block diagram form.
Figure 25: DSP implementation of SOLA algorithm |
