Arrr, distortion, she be a harsh mistress
I am working on the design for an amp, and had a few ideas about how to reduce distortion to what would hopefully be a bare minimum.
First off, the amp will be built around a PC, and will initially be tailored towards a guitar amp, however the harmonic distortion and frequency response characteristics will be highly dynamic, as the program material will pass through, or originate from the PC. Thus most any amp "sound" should be reproducible. I am going to design the data conversion circuitry and amplifier my self, thus i will have complete control over how the data is processed, and I’ll probably build it around Linux. Although i have sufficient experience in logic and data conversion design, i am fairly new to amplifier design.
Here’s what I’m thinking as far as distortion prevention goes.
Ill use the basically class AB topology discussed in endless detail in these forums before. So distortion will first be cut down on by the already established designs using negative feed back. however, rather than designing for ultimate distortion performance, ill aim for the best performance to price ratio, while providing for reasonable distortion performance, per haps something on the order of 0.1% from 20-20k. Next I’ll provide for additional distortion performance by adding more feedback with a high quality op-amp, feeding the input of the amplifier from its output, its inverting input will be tied to the output of the amplifier, and the non inverting input will be fed the signal from the DAC. I’m expecting this setup to provide for quite good performance, but have no illusions of excellence, I’ve simulated this topology and have had excellent results, but of coarse that does not mean real world results will be great.
next, because the signal is passing through the PC, which will be applying a host of filters anyway, I’ll have an input reading the output of the amplifier, so the system will be able to perform distortion analysis on it self, in which case, it can based on what it knows in terms of its own distortion characteristics, predict unwanted spectral data, and include that in its output 180 degrees out of phase, thus, with any luck canceling distortion. So for instance, if the amp produces a higher degree of distortion at double the input frequency, and the amplifier is being fed a 1 kHz signal, for each sample, the program would subtract a 2 kHz sinusoid at a 180 degree phase shift, with amplitude equal to that of the distortion, thus effectively canceling the distortion. however, one issue is that the amplifier will also distort the out of phase distortion, so it would add 4khz spectral data, so the program will also have to add a 8khz spectral data, and 32khz, and 128khz, etc. however, of coarse there will be a pint where adding more data will be of negligible advantage. Also, performance will be limited by the THD and noise floor of the feed back ADC.
also, rather than taking the feedback signal from the amp output, one could take it form a mic, thus also including any distortion induced by the speaker, or even the room. Not only will can this work for distortion cancellation, but it can also be applied to frequency response of the whole system, or be used to compensate and improve the acoustics of the room.
This is of coarse all still theoretical, id appreciate any thoughts you might have. Has anyone tried this before (I’m sure they have) and if so, what kind of results did they get?
I am working on the design for an amp, and had a few ideas about how to reduce distortion to what would hopefully be a bare minimum.
First off, the amp will be built around a PC, and will initially be tailored towards a guitar amp, however the harmonic distortion and frequency response characteristics will be highly dynamic, as the program material will pass through, or originate from the PC. Thus most any amp "sound" should be reproducible. I am going to design the data conversion circuitry and amplifier my self, thus i will have complete control over how the data is processed, and I’ll probably build it around Linux. Although i have sufficient experience in logic and data conversion design, i am fairly new to amplifier design.
Here’s what I’m thinking as far as distortion prevention goes.
Ill use the basically class AB topology discussed in endless detail in these forums before. So distortion will first be cut down on by the already established designs using negative feed back. however, rather than designing for ultimate distortion performance, ill aim for the best performance to price ratio, while providing for reasonable distortion performance, per haps something on the order of 0.1% from 20-20k. Next I’ll provide for additional distortion performance by adding more feedback with a high quality op-amp, feeding the input of the amplifier from its output, its inverting input will be tied to the output of the amplifier, and the non inverting input will be fed the signal from the DAC. I’m expecting this setup to provide for quite good performance, but have no illusions of excellence, I’ve simulated this topology and have had excellent results, but of coarse that does not mean real world results will be great.
next, because the signal is passing through the PC, which will be applying a host of filters anyway, I’ll have an input reading the output of the amplifier, so the system will be able to perform distortion analysis on it self, in which case, it can based on what it knows in terms of its own distortion characteristics, predict unwanted spectral data, and include that in its output 180 degrees out of phase, thus, with any luck canceling distortion. So for instance, if the amp produces a higher degree of distortion at double the input frequency, and the amplifier is being fed a 1 kHz signal, for each sample, the program would subtract a 2 kHz sinusoid at a 180 degree phase shift, with amplitude equal to that of the distortion, thus effectively canceling the distortion. however, one issue is that the amplifier will also distort the out of phase distortion, so it would add 4khz spectral data, so the program will also have to add a 8khz spectral data, and 32khz, and 128khz, etc. however, of coarse there will be a pint where adding more data will be of negligible advantage. Also, performance will be limited by the THD and noise floor of the feed back ADC.
also, rather than taking the feedback signal from the amp output, one could take it form a mic, thus also including any distortion induced by the speaker, or even the room. Not only will can this work for distortion cancellation, but it can also be applied to frequency response of the whole system, or be used to compensate and improve the acoustics of the room.
This is of coarse all still theoretical, id appreciate any thoughts you might have. Has anyone tried this before (I’m sure they have) and if so, what kind of results did they get?