Interesting results.
The thing here is that you're doing something the chip wasn't really intended for... making a nice waveform using logarithmic attenuators intended to do volume control is going to be a tough ask. I am surprised at how well it can actually sound.
A difference in sound volume of 1dB is 'barely perceptible' whereas the same error in a sample to sample of a digitised signal will give a very much visible error on the scope... it will probably be quite audible too but might not be sonically disastrous. Does your score do FFT analysis? That can be a better judge of how something might sound. Even a nicely constructed sample can look quite messed up once it's had all the phases messed up going through a filter. The mismatch of your tone 3 I make as less than 1dB...If I were the chip maker I would be more than happy with this as a reasonable match between channels.
A simple experiment would be to compare two normal tones on channels 1 and 3 for perceived volume with maybe a half second silence between them. I bet it will be hard to guess when it's 1 vs 3...
I can't remember the exact frequency response of the amplifier circuit in the beeb but it is fairly heavily filtered - I couldn't quite read the scales on your plots on my phone. I think maybe aim for tones around 400Hz which should pass through the filtering with their harmonics relatively unharmed.
One thing I was thinking of adding to an emulator was optional beeb speaker noise picked up from the database. It all adds to the atmosphere!
I would be interested in how the attenuators were actually constructed in silicon. It's easy in fpga and microcontrollers these days to do multiplication but really wasn't BITD.
I think the summer is probably not at issue as looking at [url=
https://www.google.com/url?sa=t&source= ... -7s7GkwhV8] this datasheet[/url] it looks, at first glance, to be a fairly standard virtual ground inverting amplifier.
Interesting stuff though... keep it coming.
The thing here is that you're doing something the chip wasn't really intended for... making a nice waveform using logarithmic attenuators intended to do volume control is going to be a tough ask. I am surprised at how well it can actually sound.
A difference in sound volume of 1dB is 'barely perceptible' whereas the same error in a sample to sample of a digitised signal will give a very much visible error on the scope... it will probably be quite audible too but might not be sonically disastrous. Does your score do FFT analysis? That can be a better judge of how something might sound. Even a nicely constructed sample can look quite messed up once it's had all the phases messed up going through a filter. The mismatch of your tone 3 I make as less than 1dB...If I were the chip maker I would be more than happy with this as a reasonable match between channels.
A simple experiment would be to compare two normal tones on channels 1 and 3 for perceived volume with maybe a half second silence between them. I bet it will be hard to guess when it's 1 vs 3...
I can't remember the exact frequency response of the amplifier circuit in the beeb but it is fairly heavily filtered - I couldn't quite read the scales on your plots on my phone. I think maybe aim for tones around 400Hz which should pass through the filtering with their harmonics relatively unharmed.
One thing I was thinking of adding to an emulator was optional beeb speaker noise picked up from the database. It all adds to the atmosphere!
I would be interested in how the attenuators were actually constructed in silicon. It's easy in fpga and microcontrollers these days to do multiplication but really wasn't BITD.
I think the summer is probably not at issue as looking at [url=
https://www.google.com/url?sa=t&source= ... -7s7GkwhV8] this datasheet[/url] it looks, at first glance, to be a fairly standard virtual ground inverting amplifier.
Interesting stuff though... keep it coming.
Statistics: Posted by dominicbeesley — Sat Apr 05, 2025 10:50 am