Guitar Effects Pitch Scaling
|
Methodology This chapter describes how an initial idea for an effect / process is developed to the stage of implementation in real time on a DSP development board. Two software packages are used in this process, Matlab and Code Explorer. The general development stages can be seen in Figure 2.
Figure 2: Development stages Matlab Matlab provides an integrated development environment that allows for the creation and testing of DSP algorithms. Therefore, this software package has been extensively used to design the various effects, as well as the pitch scaling process. Figure 3 shows the various components of Matlab.
Figure 3: Matlab development environment Section 1 above is the command driven workspace. Section 2 shows what variables are currently in memory. Section 3 is where M-files are created and edited. These can be script based or function based and allow a series of commands to be executed. Section 4 is a graphical display of a guitar signal that has previously been loaded into memory. In the case of effects development, the first stage was to investigate how the effects are generated in a digital environment. This information was found in books such as Mitra, as well as helpful websites that are listed in the bibliography. However, not all of the effect designs were from other sources. The distortion effect was designed from scratch. From the initial description / block diagram, a Matlab M-file was created to perform the effect on a wave file. These wave files were recorded from a guitar signal plugged in to the microphone connection on the PC soundcard. The sampling rate used to record is the same as the sampling rate of the ADC on the DSP development board (24kHz). Wherever possible, calculations were performed using entire vectors, rather than stepping through the vector sample by sample. While this makes the Matlab and DSP implementations radically different in many cases, it is a necessity to reduce computation time in Matlab. This is because Matlab is optimised to perform vector / matrix calculations and the alternative can be prohibitively slow. Once an effect is in the stage of coding, variables are identified that have a significant effect on the output signal. For example, the tremolo effect has a frequency variable associated with the modulating sine wave. A higher frequency gives a faster vibrato effect. Such variables will be modified by the guitarist to give a wide variety of alternative sounds. The Matlab stage is an iterative one. Modifications are made to the algorithm until suitable results are achieved. Conversion to Algebraic Instruction Set (AIS) Many Matlab functions are at a very high level of abstraction. For example, the function XCORR performs correlation between two vectors in one command. To be able to implement such functions on a DSP, they need to be broken down into AIS, which is what is used to program the DSP development board. AIS is an improvement over mnemonic assembler, as you can write the code with mathematical equations, rather than specific operations. However, it is still at a low level of abstraction compared with Matlab. The conversion process from Matlab M-file script to AIS needed to take into account that the DSP only has 10kB of available memory. The speed efficiency of an algorithm at times had to be reduced to ensure that less memory was required. Once the AIS code was developed, it was compiled into an object file that can be uploaded into the DSP development board. Development Board Debugging The development board is designed to allow for a wide range of applications to be tested on a real DSP, rather than using an emulator that runs on a PC, that might not be able to achieve real time testing. The board has an Analogue Interface Chip (AIC), and serial port connection to the host PC. Figure 4 on the following page shows the typical workspace of the software program Code Explorer. This program allows a developer access to all of the DSP’s memory (Figure 4, section 2) and registers (section 4), and enables debugging of a program through instruction stepping and breakpoints (section 1, program disassembly). It is also possible to graph areas of the memory as shown in section 3. An electric guitar is connected to the input of the AIC, and a speaker to the output. The effects and pitch scaling processes were tested in real time to ensure that conversion process worked correctly. The delay was also examined to make sure that it was not audible. As with the Matlab section, an iterative approach was taken to code, test, and modify until a suitable result was achieved.
Figure 4: Code Explorer interface with DSP development board
|
