Filter and sampling frequency

4 posts / 0 new
Last post
Morgado
Morgado's picture
Filter and sampling frequency

I was reading a paper and I saw the following phrase in "Materials and Methods":
"Whole-cell currents (potassium currents, Kv) were filtered at 5 KHz and sampled at 25 kHz".
In another site I saw the following protocol:
"Whole-cell currents were filtered at 1 kHz and sampled at 40 kHz".
What is the filter and sampling frequency?
How should we choose the filter and sampling frequency?
Thanks in advance for any insight into these questions.
Morgado
 

Fraser Moss
Fraser Moss's picture
Normal

Normal
0

false
false
false

EN-US
X-NONE
X-NONE

MicrosoftInternetExplorer4

When recording your potassium currents you select an acquisition frequency that sets the interval time for the acquisition of each data point in the trace.  The amplifier records an analogue signal from the headstage but has to convert this into a digital output for storage on the computer.  The interval between the digitized data points sets the resolution of the acquired signal.  The experimenter needs to select an acquisition sampling rate that reasonably represents the raw analogue signal.  You cannot sample at too high a frequency as this will just record more data points, but will also begin to include intrinsic noise and other noise in your signal the faster you sample and furthermore the large size of the recorded files just take up more and more disk space.  However, if you sample at too low a frequency then the digital signal will be distorted and will not be an accurate representation of the analogue signal.  You need to select a sampling rate that is sufficient to detect the all critical events you intend to record such as channel opening, closing, inactivation and/or desensitization without distortion.  Ideally the sampling rate is at least 100 times faster than the event to be recorded. This is the “sampling frequency” that you see quoted in papers.

The accompanying intrinsic noise can be filtered.  An electrophysiological circuit has intrinsic noise generated by the components of the system. You can take steps to minimize the generation of intrinsic noise on your rig, but most times it is necessary to process the recorded signals in order to extract the useful signal from the noise.  This is filtering
There is a very important principle that everyone should know about filtering and that is that by definition it leads to the loss of data.
It is always best to sample and store data in its rawest possible form that allows analysis of the signal, and “offline filtering” can be performed after the acquisition of the data so that the maximum data is recorded before the signal is extracted without the loss of information.  In reality this is not usually an option because if the noise levels are high, the amplitude resolution of the stored data is not good enough low to allow efficient offline filtering of the real signal (low signal-to-noise).
The analogue gain of the Patch clamp signal is turned up as high as possible to obtain good signal to noise for the  analogue to digital conversion, but the signals usually still needs some filtering before they are stored.  The general rule is to acquire the highest resolution signal that you can including the noise (ie as many data points as possible) and not record the prettiest looking trace with a low noise that you can visualize at the time of acquisition.
The most common filter used in biological circuits is the low-pass filter.  The filter settings you are applying on the amplifier are low pass filters and will cut off all frequencies above that set by the operator.  The cut-off frequency of a low-pass filter is defined as the frequency where the power of the original signal is dampened by -3 decibels (dB) or about half its original strength.
The frequency that you set for the low-pass filter is usually ½ the sampling frequency (e.g 2khz sampling 1khz filtering) but oftentimes experimenters will use 1/5 the sampling rate (25khz sampling, 5khz filtering).  As long as the filtering does not mask any of the physiological events being recorded then it is OK. The sampling and filtering rates vary a lot between channel to channel, and which event is being recorded (e.g. channel opening rate, steady state current, inactivation rates etc). 
Again just make sure you DO NOT OVER FILTER the recording at acquisition.  You can always apply additional offline Gaussian filters after the data is stored to clean up the trace.  However, you cannot put data back in that was lost due to over-filtering.
for more information on sampling and filtering read the Axon Guide at http://www.moleculardevices.com/pdfs/Axon_Guide.pdf

lazy
lazy's picture
Thanks for great information.

Thanks for great information.
Just I like to know the ideal sampling rate and filtering for standard hERG recording:
My voltage protocol is 1 sec at -80 mV, 2 sec at +20 mV , 4 sec at -50 mV, and 2 sec at -80 mV.
The trace length is thus 9 sec. Also the response recorded is not something very rapid (I believe?). I tried 3 different combinations of sampling and filtering.
1: 10 kHz sampling and 3 kHz filtering (signals lower than 3 kHz can pass the filter?)
2: 1 kHz sampling and 3 kHz filtering (signals lower than 3 kHz can pass the filter?)
3: 1 kHz sampling and 0.5 kHz filtering (signals lower than 0.5 kHz can pass the filter?)
From my understanding, (3) seems to be reasonable, while I do not know if (2) is wrong or not (maybe S/N ratio is relatively low). I noticed bi-phasic capacity current in (1), so I do not think this is good combination.
Thanks for your comments in advance.
 

Fraser Moss
Fraser Moss's picture
1. would work, but your don't

1. would work, but your don't need to sample that fast for hERG unless you are trying to record steady state inactivation properties. Also you shouldn't see any capacitance current if you are applying Capacitance and series resistance compensation correctly no matter the sampling/filteing frequencies.
2. Would not filter anything because you're sampling slower than you are filtering
3. is  probably ok, especially if your currents are greater than 500 pA (good enough S/N).
I have a similar 10 sec hERG protocol in which I smaple at 2kHz and filter at 1 Khz and this works just fine for acquiring dose response curves and I-V relationships