ECG amplifier noise issue

[scottlc]

I have built an ECG amplifier circuit using an INA128 instrumentation amplifier with a driven right leg. The circuit is powered by two PP9 batteries for +/- 9V operation and is assembled on strip-board. This circuit suffers from a large amount of broadband noise which swamps the fine detail of the P and T waves. A full circuit diagram, photo and sample data is attached. The data is sampled at 1600 Sa/s and is in volts.

Measurements are made using 3M Red Dot 2330 diagnostic electrodes placed on the wrists with the right leg drive attached to the right ankle. I've also tried using LF347 and OP471 quad op-amps, though they produce roughly equivalent noise output. INA128 gain being set to 10, 30 or 50 seems to have no effect on the signal to noise ratio. Noise is minimal with all three probes connected together (constant signal, within ADC quantisation step). Noise is also minimal when the right leg drive is attached to the right leg and two electrodes are placed side-by-side on the left wrist. I've also tried reducing the value of the resistor on the integrator feeding the INA128's reference terminal with no effect. I've also tried swapping the centre tap of the INA128's gain resistor (common mode) to ground to effectively disable the right leg drive, but this has no effect.

Signal is consistent when measured with a Gwinstek 200MHz digital scope, National Instruments USB-6009 14-bit device, and with the 12-bit ADC input of an ATMEGA168 microcontroller. Noise is consistent when measured away from fluorescent lighting and when completely isolated from mains power, and also when powered from a lab power supply instead of battery.

On the positive side there is no 50 Hz mains interference visible above the noise floor.

At this stage I'm completely out of ideas. Any insight or suggestions would be greatly appreciated.

 

[chuckey]

I don't understand your circuit around the two 470 ohm resistors, looking at the top amps, their gain is 25k/470, but only if the 470s go to low impedance path to earth but they go to the output of the other amp via another 470 +25K and to the input of the bottom amp. Is there any stage gain here?
Frank

 

[ZekeR]

Out of curiosity, what does the spectrum of the noise look like? (If your oscilloscope has an FFT function, you should be able to obtain a spectral plot). I somewhat doubt that it is white noise; I have a hunch that you'll see spectral peaks at some frequency (perhaps in the kHz range, probably due to a switch-mode power supply nearby). If this is indeed the problem, it is possible to filter it out.

Hope this helps.

- - - Updated - - -

chuckey, the circuit behaves as an instrumentation amplifier with gain 1+2*25k/(470+470)=54. The bottom amplifier compares the center tap of the "gain" resistor with GND, and then servos the human's potential (through the right leg drive) to keep LA and RA centered around GND. This practice is called "common-mode feedback" (CMFB), which is fairly common in differential circuits to keep the inputs within the normal operating range.

 

[scottlc]

There is no obvious structure to the noise as far as I can tell. Attached is a small amount of data in the silent region between beats and the corresponding FFT (Hann window, dB scale, frequency in Hz).

 

[ZekeR]

Hmm, I don't see anything notable here (I imagine that if there was any noise with a sharp spectral peak, it'd get aliased down into baseband and you would see it). A possibility is that there's a noise source with lots of jitter (on the order of multiple hundreds of Hz), and therefore the aliased noise doesn't see a spectral peak. It might be worth it to take a closer look at the spectrum at higher frequencies. (I suggest using your Instek, if it has an FFT).

I'm still leaning toward the idea that it's noise injection from nearby EMI sources. AFAIK, the solution would be
a) Remove the noise source
b) Add lots of shielding
c) Use multiple ECG probes and find the correlation between them

 

[scottlc]

I've confirmed with the scope that there is no noise beyond my sampling bandwidth. Adding two cascaded 190 Hz low pass filters with intermediate unity buffering has no noticable effect on the noise. Probing at the output from the INA128 shows a lot of noise at low frequency and beyond, much of which is removed by the filtering, but the noise that I'm seeing in the output appears to be also present here. Again, this noise seems to be coupling to the circuit through the electrodes since the circuit behaves well when all three electrodes are tied together.

I've been performing most of the work on this circuit in a research lab, but the noise was exactly the same when measured at home away from mains sockets with no lighting on and sampling using a laptop running on battery.

I don't have a lot of experience designing analog circuits so I'm curious about the possibility that this is down to some deficiency in the circuit design.

 

[chuckey]

get hold of some capacitors , say 2 MF, starting at the output, connect them across your signal path and work towards the input. Hopefully at some stage the noise will go or at least be severely attenuated. Lo and behold, you have found the noisy stage!
Frank

 

[ZekeR]

Well, I'm fresh out of ideas at the moment. Since the noise goes away when you connect the probes together, that would indicate:
a) The noise isn't due to feedback loop instability
b) The noise isn't due to the selection of op-amps or resistor values
c) The noise is not otherwise contributed by the circuit or injected into the circuit.

This indicates that the noise must be contributed by the human's body, as a potential across his/her arms. This leaves two possibilities:
a) The human is picking up electromagnetic interference from the laptop or something
b) The noise is the result of other signals being carried through the human's body (brain waves, digestion, flexing muscles to impress girls, etc.)

I could be wrong, but it seems to me the only steps that can be taken would be:
a) Increase shielding to prevent EMI
b) Use multiple probes and find the covariance between their signals, to weed out other bioelectrical impulses
c) Use large probe pads to average out local bioelectrical impulses

I'm curious to find out what ends up working for you, so keep us posted.

 

[scottlc]

I've isolated the noise to be EMG interference. If the subject is lying down for the measurement, still and relaxed, then there is considerably less noise than sitting. I'm getting pretty clean traces now. Example attached (unfiltered, some 50 Hz mains hum present).

 

[ZekeR]

That's a nice looking ECG. Good job. =)

 

[scottlc]

I've also had a quick attempt at wavelet filtering with pretty good results.