CW Audio Filter: Narrow Passband, Ringing and DSP

Anyone who has copied a weak CW signal on a crowded band knows the experience. The note is there, but it is buried in hiss, and a strong station a few hundred hertz away keeps tapping on the edge of your attention. A CW audio filter is the simplest way to improve this situation. It lets through only a narrow slice of the audio around the pitch of the wanted tone, and removes the noise and the neighbours outside it. Used correctly, such a filter can make a barely audible signal readable. Used incorrectly, it makes the dits and dahs sound like soft bells, with a ringing tail that smears the code. This guide explains the theory in plain terms, compares the kinds of filters, shows where they work and where they cannot, and gives settings you can try tonight.

Why a Narrow Filter Helps

Noise is spread evenly across the frequencies, so the noise power that reaches your ear is proportional to the bandwidth of the receiver chain. A CW signal, however, occupies only a very small bandwidth, roughly a hundred hertz or less at normal speeds. A receiver with an SSB filter of 2.4 kilohertz passes twenty times more noise than the signal needs. If you narrow the passband from 2,400 hertz to 400 hertz, the noise power falls by a factor of six, which is 7.8 decibels, while the signal passes unchanged. In practical terms, a signal that was 3 dB below the noise now stands 5 dB above it. The comparison of the decibel values is explained in the article on decibels in ham radio.

The second advantage is the rejection of neighbouring signals. If a station transmits 300 hertz above your frequency, it is normally audible as a second tone in the headphones. A filter centred on your pitch removes it, so that you hear only one signal at a time. This is the audio equivalent of a narrow crystal filter, and it is one reason why experienced CW operators seem to hear signals that others cannot.

The Price: Filter Q and Ringing

A narrow filter cannot react instantly. When a tone starts, the output of a resonant filter builds up over a time that is inversely proportional to its bandwidth. When the tone stops, the filter keeps oscillating for a while, which is called ringing. The effect is described by the filter’s quality factor Q, the centre frequency divided by the bandwidth (see Q factor). A filter centred at 700 hertz with a bandwidth of 100 hertz has a Q of 7. A bandwidth of 50 hertz gives a Q of 14. The ringing time is of the order of the reciprocal of the bandwidth: about 10 milliseconds for 100 hertz and 20 milliseconds for 50 hertz.

At 20 words per minute a dit lasts 60 milliseconds, so a ringing tail of 10 to 20 milliseconds is acceptable. At 35 words per minute a dit lasts only 34 milliseconds, and the same tail begins to fill the gap between the elements. The result is a note that rings, the spaces between characters seem shorter, and letters are harder to separate. This is why operators who work fast CW use wider filters than slow-speed operators, and why the best bandwidth depends on the speed. A useful rule is that the bandwidth in hertz should be at least about twice the speed in words per minute, and more for comfort.

Kinds of Audio Filters

The first kind is the classical active analog filter, built from operational amplifiers, resistors and capacitors. A band-pass section, for example in the Sallen-Key or multiple-feedback arrangement, provides a Q of up to ten with a single stage, and cascaded stages create a steeper response. Analog filters are simple, cheap and have no delay, but their centre frequency is set by components and drifts a little with temperature, and a high Q needs good parts. Many homebrew receivers and the add-on audio filters sold for the shack have this design.

The second kind is the switched-capacitor filter, built around a chip that uses a clock signal to set its characteristics. The advantage is that the centre frequency follows the clock, so it can be tuned with a knob, and the bandwidth is selectable. The disadvantage is the clock noise, which appears as a faint whine in the audio unless it is filtered afterwards. The third and today most common kind is the digital filter in a DSP chip. It can create a band-pass with a steepness that analog circuits cannot match, with the centre frequency and the width set from the front panel, and it can be combined with a noise reduction function, a notch and an automatic peak filter. The price is a small delay and, in a poorly designed unit, quantisation noise.

Peak Filters and Band-Pass Filters

A band-pass filter attenuates everything outside its passband and passes the band without changing it. A peak filter, also called a peaking or boost filter, raises the level of the frequencies near its centre while leaving the rest almost unchanged. For CW it gives the impression of a narrow filter while the surrounding noise is only reduced, not eliminated. Peak filters ring less than a band-pass of the same width, because they do not remove the surroundings completely. Many transceivers offer a filter called APF, the audio peak filter, which is very pleasant for tracking a weak station without losing the wider spectrum. Good practice is to use the peak when you want to hear the neighbourhood and the band-pass when you must resist interference.

Where the Filter Sits in the Chain

The position of the filter matters as much as its width. An audio filter works after the detector and after the automatic gain control, so it cannot stop a strong signal from reducing the gain of the receiver. If a loud station 1 kilohertz away captures the AGC, the weak station you want falls in level together with everything else, even though the audio filter removes the tone of the strong one. For this reason the first line of defence is the IF filter or the DSP filter ahead of the AGC detector, as described in the article on receive filter bandwidth and roofing filters, and the AGC behaviour is explained in the guide to AGC. The audio filter is the last stage that cleans up what remains.

A practical test shows the difference. Listen to a weak CW signal with a strong signal 800 hertz away. If the weak one pumps up and down in volume each time the strong one keys, the AGC is the cause and an audio filter will not help; switch to a narrower IF filter, reduce the RF gain or shorten the AGC time constant. If both signals are of comparable level and you merely hear two tones, the audio filter is the right tool.

Setting Up a Filter for CW

Begin by tuning the transceiver so that the pitch of the incoming tone matches the centre of your audio filter. Most radios use a sidetone or CW pitch setting, usually between 500 and 800 hertz. Set it to a value that is comfortable for your ears, and zero-beat the signals to it, as explained in the article on how to zero beat a CW signal. A mistuned signal falls outside the narrow filter, and it disappears, which is the usual reason why a new user of a narrow filter says that the signals vanish.

Start with a bandwidth of 400 to 500 hertz and narrow it in steps. Stop narrowing when the signal begins to ring or the characters lose their rhythm. If you are working a contest at 30 words per minute, you may end up at 400 hertz, and when you listen to a slow weak station, a width of 150 hertz may be fine. Many operators switch between two settings during the session, and a fast switch makes the choice easy. Keep the volume at a reasonable level: a narrow filter with a high volume leads to listening fatigue as the ear hears the ringing tail of the noise.

Choosing the Pitch

The centre of the filter is also the pitch you will listen to for hours, so choose it with care. Many operators like a note between 600 and 700 hertz, but there is no rule, and ears differ. People with some high-frequency hearing loss often prefer a lower note of 400 to 500 hertz, while others find a higher pitch around 800 hertz cleaner and easier to separate from low-frequency hum and rumble. Try different values during an evening and notice which one tires you least after an hour.

A second reason to think about the pitch is the sidetone of your own keying. The sidetone should match the receive pitch so that your transmitted signal is exactly on the frequency where you listen. If the two differ, you zero-beat by ear incorrectly and may call a station slightly off its frequency, which in a pileup costs you the contact. With a narrow filter this is even more important, since a signal that is 100 hertz off may fall at the edge of the passband and sound weak. Set the sidetone and the receive pitch to the same value, and keep them there.

Finally, remember that headphones change the picture. Closed headphones suppress the room noise and make a narrow filter more effective, while a loudspeaker in a noisy room hides its benefit. A comfortable pair of headphones is often a better investment for CW than another filter.

Noise Reduction Is Not the Same as a Filter

A filter reduces the noise outside its band. A noise reduction function, in contrast, tries to identify the noise inside the band by its statistical properties and subtract it. The two are complementary, and the effect of the noise processing on CW depends on the algorithm. Aggressive settings can make the dits sound like short pops and cause a watery quality of the tone, so use them at a moderate level, as described in the article on noise blanker and noise reduction. For a weak signal, first narrow the filter, then add a small amount of noise reduction, and compare against the off position often.

Filters for Decoding Software and Skimmers

If you decode CW with a program, an audio filter helps there too. Decoders look for a tone and a rhythm, and a clean narrow signal improves the result. The skimmer programs described in the article on CW Skimmer and the Reverse Beacon Network use sophisticated filtering of their own and need no help, but a simple decoder on a phone responds well to a clean tone. Do not rely on a decoder to learn, though: the brain is a better decoder than any software, and the filter is there to help the brain.

The Bottom Line

A CW audio filter passes only the narrow slice of audio around the pitch of the wanted tone. Narrowing the passband from 2,400 hertz to 400 hertz reduces the noise power by about 7.8 decibels, but a high Q makes the filter ring, which blurs the code at high speeds. Choose the bandwidth to match the speed, at least twice the speed in words per minute, and go narrower only for slow, weak signals. Analog, switched-capacitor and DSP filters all do the job, with DSP offering the most flexible control. Remember that an audio filter works after the AGC, so rely on the IF or DSP filter before the AGC to handle strong neighbours, and tune each signal exactly to the filter centre.

Frequently Asked Questions

What bandwidth is best for a CW audio filter?

For general CW, 300 to 500 hertz. Go down to 100 to 150 hertz for weak signals at slow speed, and use a wider setting for fast speeds and contests.

Why does a narrow filter make the code ring?

A resonant filter takes time to build up and decay, in proportion to the reciprocal of its bandwidth. At high Q the decay tail fills the spaces between dits and dahs.

Does an audio filter replace a roofing filter?

No. It works after the AGC, so a strong nearby signal can still reduce the gain of the receiver. A narrow IF or DSP filter ahead of the AGC deals with that.

What is the difference between a band-pass and a peak filter?

A band-pass removes everything outside its band. A peak filter raises the level around its centre and leaves the rest nearly unchanged, so it rings less and keeps some of the surrounding sound.

Why do signals disappear when I narrow the filter?

The tone is probably not at the filter’s centre. Tune the signal so that its pitch matches the filter centre, usually the CW pitch setting of the radio.

Is a DSP filter better than an analog one?

Usually it is more flexible and steeper, with easily adjustable width and centre, although a poor implementation may add delay or quantisation noise.