20.07.06 Added a second mode to the ECG firmware enabling it to store the measured data with 200Hz. Unfortunately it isn't then any more possible to store the statistical error for each measurement, but each second an average error which rather represents a quite conservative guess for te actual error is stored. Regarding this error the signal looks much better than the error is proposing. Using the 200Hz measurements the QRS complex is by far better obtained, see 100Hz and 200Hz.
19.07.06 After some debugging, the first real 3 channel measurement with the new board: see here As usual 10000 on the Voltage axis is roughly corresponding to 1mV.
10.07.06 current status of soldering the 3 channel version : click

Havn't mentioned it here: The 'cent-electrodes' are getting bad after using them about a week, so either way you have to sand them or replace them by new ones. Buying real resting ecg electrodes with some sort of AgCl gel in them might be worth a try too, you get a clean signal and low amplitudes of the usual 50Hz part.
10.07.06 current status of soldering the 3 channel version : click
22.06.06 A talk on a tour at BOSCH organized by the university about image processing in quality control using methods similar to the 'eigenfaces' method made me think about analysing the EKG data using a similar algorithm. Let's see how much better results than just comparing to the averages one might get. Normalizing and interpolating pulse periods has already been done, so diagonalizing the covariance matrix didn't cost much effort ... I guess interpretation will :-)

Main problems with stability of the signal must have been caused by the alumium foil electrodes, the '2cent'-electrodes work way better, check out this. The really simple onboard pulserate detection is working now due to the better signal.
Hard- and Software has been extended to also monitor the power supply voltage, since I observed that old batteries were causing trouble. To save space only time differences are stored on the mmc card and once in a second also the status data like the supply voltage is stored to the memory card.
I'm planning to compare running the AD converter at 115 and 230kHz, where the latter would be outside the specification for a 'good' 10Bit ADC, but since we might have more samples the error of the average might be better - that's left to be checked
Postprocessing on the PC itself has been extended by pulserate detection and by preliminary averaging over multible periods. Idea is to 'normalize' each period of the ECG to some length and then categorize them onto a 2D plane using a SOM and see if certain features can be recognized and/or faulty parts can be recognized. Another idea is to train a relatively simple feedforward net for QRS detection.

09.06.06 Malfunction of my primary 'signal source' has been fixed, hopefully ... *g*