From: Joerg Hansmann (info_at_jhansmann.de)
Date: 2001-05-20 18:36:58
Hi,
----- Original Message -----
From: Brian Gr <brian_gr_at_hotmail.com>
To: <buildcheapeeg_at_yahoogroups.com>
Sent: Sunday, May 20, 2001 6:34 PM
Subject: [buildcheapeeg] Is Sound Card feasible?
Hi,
First post to this list. In listening to the discussion about the sound card as a potential input device in abstract discussions, they seem to have some merit. But, could someone explain how they see a sound card being usedin practical terms? I've heard two methods described, a VCO method and an A/D converter method, but i see serious problems with both methods.
There was a third method: Time multiplexing.
With the VCO method, I don't see how high enough signal resolution could be obtained at working sample rates.
Lets assume the sound card can sample the isolated VCO output at 40Khz (the top soundcard sample rate is somewhere around there) Let's assume the VCO is set for 2Khz = -100uV, 4Khz = 0V and 8KHz = +100uV.
Why do you assume a nonlinear VCO characteristic ?
Normally VCOs are linear (e.g. CD4046BC has High VCO linearity: 1% (typ.))
Now, how quickly do we need to sample the data stream? Let's say we want 200 EEG samples per second. On a continious basis, in 1/200 of a second, wemust determine the VCO frequency (for each channel) to translate this backto the VCO input (ie a representation of the EEG signal) Sampling the VCO output at 40Khz, means we will have 200 samples of the 4Khz VCO output at 200 samples per second. We would then count the number of zero crossings in these 200 samples to ultimately determine VCO frequency. 4KHz VCO output (0V signal) in 1/200 second = a count of 20. At 2KHz, (-100uV), count in 1/200 second = a count of 10. At 8KHz, the count is 40. So, what's the resolution?
Between -100uV and 0 we have 10 discrete steps. Between 0 and 100uV we have 20 steps. We have better resolution on the positive part of the signal, but the resolution is still about 4 bits, not useable for EEG.
The resolution could be enhanced to a few bits more with e.g linear regression around the zero transition of the sampled sine-wave.
And, we've got to process a data stream from the sound card at 40Khz, a daunting task in the midst of processing the EEG data and driving a graphicdisplay.
ACK.
It would be nice if our frequency counter(s) had 1 second to produce a sample, because the resolution would be somewhere around 10 bits. But, an EEGsystem running at 1 sample per second would have serious problems. Is my understanding of the signal demodulation totally wrong....I have been knownto miss the forest for the trees?
No. You see the problems quite clearly.
With the A/D method, the suggestion is to modify the input circuitry of the sound card to extend the low frequency limit. Might be problems here. Surface mount components on the soundcard, integrated sound cards like on a laptop, unique modifications for every sound card or motherboard, calibration, who's got the particular sound card schematic and expertise to modify the low frequency limit of each sound card? The safety issue must also be addressed by optically isolating linear signal channel(s) between the EEG amplifiers and the sound card input; more costly than isolating a multiplexed (ie one channel) digital stream. This method would potentially limit the number of computers that coould run this system.
I would see the above as making the COM port method, with a small AD and processor desireable and the path of least resistance. Unloads PC considerably (no multi channel FM demodulation), COM port has more than enough bandwidth for a couple of channels at several hundred samples per second.
I agree.
Regards,
Joerg
This archive was generated by hypermail 2.1.4 : 2002-07-27 12:28:30 BST