A Hum Filter

The filter is intended for use when listening to VLF radio signals in the presence of foreground or background hum.

The program features:

• Multiple delay stages to obtain a very sharp notch which reduces audible coloration of the signal.
• Period space tracking of the line frequency, which accurately maintains the harmonics in the deepest point of the notch.

This filter copies the sound card input to the output, tenaciously following the line frequency hum with a deep notch. When properly set up, the harmonic components of foreground hum should almost all the time be completely inaudible at 16 bits dynamic range.

Anyone is welcome to do what they want with this code. You will most likely want to use this program to try out the filter, and then import the filter functions into your own software.

Description

Sharpness of the multiple stage notch necessitates accurate tracking of the varying line frequency (typically wandering by up to +/- 0.5%) to ensure that the line harmonics remain in the deepest point of the notches. This can be done by using the delay buffer and summing function to generate a 3-point periodgram, which is used to steer the delay time towards that which gives maximum output from the summing function, and therefore the deepest notch.

The depth of the notches and the tightness of the period lock enable the harmonic components of foreground hum to be removed almost completely. The program uses linear interpolation when drawing samples from the delay buffer, which improves the accuracy of the notch for those delay times which are not represented by an integer number of samples.

Caveats and Tips

• Delay tracking doesn't work for less than two stages.
• This type of filtering will only remove harmonic components of the foreground hum. Non-harmonic components can be a real problem, especially with signals from H-field antennas. Other techniques must be used to deal with these.
• It is vital to ensure that foreground hum does not clip anywhere in the receive path upstream of the soundcard, and that the soundcard mixer gain setting is such that the signal does not go out of range on the A/D. The program reports the peak input level so that you can monitor this. If overloading is allowed to occur, the hum cross modulates with the noise to produce a broad spectrum of interference which cannot be removed.
• When dealing with foreground hum, 16 bit mode is preferable. If using 8 bit mode and the hum is 10 times louder than the signals, the remaining dynamic range after filtering is only 5 bits worth.
• For simplicity, the program makes no attempt to customise the buffering on the soundcard i/o. As a consequence, there may be a considerable time delay between input and output.
• In the event that the input hum is so weak that the filter can't recognise it, it will probably slew to one of its end stops and sit there harmlessly.
• When running with 10 or more stages, the filter can lock onto one of the dips in the passband ripple. This occurs if the filter is started from cold and the line frequency happens to be a long way from its nominal value. Usually after a few seconds the filter finds the correct notch.
• The ultimate depth of the notch is determined by the accuracy with which the outputs of the delay stages can be summed, which improves with increasing sample rate and increasing number of stages.
• There isn't much audible benefit in going beyond 4 or 5 stages.
• This filter is intended to improve the audible quality when listening to VLF signals. It generates quite a bit of passband ripple which can sometimes be a nuisance when taking signal measurements. For example when receiving ELF signals below the line frequency, a straightforward notch filter for the line frequency together with a low pass filter for the harmonics is likely to be much more satisfactory.
• This demo program as it stands, only works on Pentium processors when running with 16 bit samples. Users of big endian machines will have to alter the code slightly.

CPU Consumption

  samples/sec    stages     usage

     8018            2        2%
    18092            2        5%
    50400            2       16%
     
     8018            4        4%
    18092            4        9%
    50400            4       26%
     
     8018            6        6% 
    18092            6       13%
    50400            6       38%

     8018           10       11%
    18092           10       21%
    50400           10       57%

Installation

If you have 60Hz line frequency, edit humfilt.c and change NOMINAL_DELAY from 20.0 to 16.6666. If your sound device is not /dev/dsp then edit the value of DEVICE. The other default settings should be fine as they are for most situations.

On Linux systems, compile the program with,

  cd humfilt-1.3
  make

Running the program

   ./humfilt -16 -r 18092 -s 4

The program may take a few seconds to lock onto the hum.

The full set of command line options are:

• -8 Run in 8 bit mode.
• -16 Run in 16 bit mode.
• -r rate Try to set rate samples per second.
• -t cycle Revise the filter delay every cycle seconds.
• -s stages Use stages number of delay sections.

The filter can be switched in and out by signalling the program. To turn the filter off, use

  killall -USR1 humfilt

  killall -USR2 humfilt

My Experiences

Some example VLF sounds which have passed through this filter are:

• aur.100203a.mp3, 390kB, 100 seconds. Auroral 'dawn chorus', present to some extent about 1 morning in every 5 or 6 at my 54 deg north location. Sometimes present in the afternoons and evenings.
• sigs.110203.mp3, 155kB, 40 seconds. Background activity during the late afternoon. Very few sferics and a couple of auroral risers can be heard. There is also some non-harmonic power line interference. I switched the hum filter out for a few seconds during this sample to demonstrate the raw signal.