HF Band Passive Noise Cancellation
Gwyn Griffiths G3ZIL describes the construction and use of three simple modules
Gwyn Griffiths G3ZIL describes the construction and use of three simple modules that, together, may reduce your HF band noise.
Like many, my HF band reception is affected by noise. Nigel Squibb G4HZX and I described a number of measures that reduced our noise in PW October 2017 but it’s an ongoing challenge. One method that I thought I should try was noise cancellation prior to the antenna socket of the receiver. Paul Beaumont G7VAK wrote an interesting and informative PW article on the topic in March 2014. There are several commercial devices available, including the MFJ-1026, the NCC-2 from DX Engineering and the ANC-4 from Timewave. This method of noise cancellation needs an auxiliary or ‘noise’ antenna, a means to match the amplitudes of the noise from the noise and main antennas and a method to alter the phase so that at a combiner the two noise components are 180˚ out of phase.
Paul’s article and online reviews of the commercial devices provide compelling evidence that the method can work. However, getting good results can be fiddly and much depends on your own noise environment and the noise antenna. My approach was to see if I could implement the method and achieve useful results with three aims in mind:
- A completely passive solution, avoiding broadband preamplifiers that might degrade receiver performance.
- To leave me with useful modules if the method did not work in my own situation.
- To have just two controls (amplitude and phase) and for those to be of fixed known steps and to have definite meaning.
The Modular Passive Approach
Three modules are needed, whose construction is described in this article: a broadband zero phase passive combiner, a switched delay line and a switched attenuator. The block diagram, Fig. 1, shows how they are connected together. Also shown are the interconnecting coaxial cables (RG58C/U), their lengths, and the positions of the two antennas. In my case both are shallow inverted-V dipoles. The cable lengths are not important; they are what they need to be for the particular antenna arrangement and the location of the passive combiner. The cable length on my noise antenna is longer than it need be in this arrangement because I wanted to keep the same length while I tested various noise antenna positions.
The main antenna is as far from the house as possible. Both dipoles run in the same orientation parallel to the back wall of the house. The attenuator is placed in the noise antenna feed because its noise output has been found to be equal to, or greater than, the main antenna. The switched delay is in the feed from the main antenna so as to ensure 180˚ phase shift given the cable lengths, the antenna separation and that the noise to be cancelled comes from the direction of the house. As the set-up and cable lengths may need to be different at different locations you may find that if you cannot find a null, you may need to connect the delay line in the noise antenna feed.
The Passive Combiner
The passive combiner module is built in to a 92 x 38 x 31mm aluminium diecast box (type 1590A, often known as a ‘stomp box’). The circuit chosen is that of two autotransformers and a single resistor, Fig. 2. I used a 0.25W 1% metal-film resistor in this micropower application. Both transformers are wound with 22swg enamelled wire. T2 is centre-tapped, and because equal and opposite currents flow in the transformer and the resistor there should be high isolation between the two ports A and B. With 50Ω impedances connected to ports A and B, the impedance at the centre tap is 25Ω, hence T1 is a 25 to 50Ω matching transformer with a turns ratio of 1.4:1, that is, the tap is at five turns from the earthed end. While the theoretical loss is 3dB, the actual loss was no more than 4.3dB over the HF bands. The module can also be used as a splitter by applying the input signal to port S. The arrangement within the box is shown in Fig. 3, with short leads and the two toroids at right angles to achieve good isolation.
The Switched Delay
Commercial noise cancellers and that described in PW by Paul G7VAK use a transformer-driven bridge network with a variable resistor to set the phase shift over a 180˚ range. I took the alternative approach of using coiled lengths of coaxial cable to provide fixed delays. In part, this was to take a slightly different approach but it does make setting repeatable and certain, which can help with interpreting the results. With six cable lengths my aim was to span up to 76ns (71ns being 180˚ on my band of main interest, 40m, with a little extra in hand) in steps of (mostly) 1ns, that is, in phase increments of about 2.5˚ on 40m. Fig. 4 shows the simple circuit with the cable lengths and the measured delays while Fig. 5 shows the completed unit. The cable length needed is simply the speed of light in vacuum multiplied by the required delay time and the cable velocity factor, vf. For RG58C/U vf is 0.66. You can use other cable but do recalculate the lengths with the appropriate vf.
The switched delay line was housed in a simple bent aluminium sheet U-shaped box, with short cables to two front-mounted sockets. One additional switched section with 14.2m of cable would extend operation to 80m. From the results below, 1ns resolution should mean that the unit is usable to 10m.
As an aside, the front panel lettering was done by inkjet printing onto Letraset Safmat self-adhesive film. After printing, the sheet was sprayed with several coats of clear acrylic lacquer and left overnight. The biggest challenge in applying to the panel is avoiding air bubbles. Those that do form can be pricked with a fine needlework pin and pressed out.
The attenuator is perhaps the module you may already have to hand. There are a number of inexpensive commercial step attenuators available and one of these could be used but building your own is a simple job. A range of 0-15dB in 1dB steps is likely to prove more than adequate, given that too much disparity between the two antennas may compromise noise cancellation. The simple 50Ω four-stage switched attenuator of Fig. 6 using T sections with standard E24 value resistors is sufficiently accurate. I used 1% 0.25W metal film resistors for 11Ω and above but could only find 0.25W carbon film 5% for the 2.7 and 5.6Ω. Nevertheless, from 100kHz to 100MHz the measured attenuation was no more than 0.2dB from the correct value for all settings. 2.5mm copper wire was used for the ground bus-bar and 1mm copper for the interconnections. Another aluminium diecast box type 1590A provides the housing, Fig. 7.
I tested the three modules in combination on the bench using an amplified zener noise source and a spectrum analyser connected to port S of the combiner. The total cable length in the attenuator path was 2.45m and 2.58m for the switched delay path. At the test frequency of 7.04MHz you would expect a setting of 71ns to give 180˚ phase shift. With the approximate attenuation in RG58C/U cable of 0.036dB/m at 7MHz, the calculated attenuation in the delay path would be 0.5dB but perhaps a few tenths higher due to switch losses. Therefore, I’d expect to have to set the attenuator to 1dB to give equal noise levels at the two ports of the combiner.
Fig. 8 is a 3D plot showing a maximum of 36dB noise reduction at settings of 72ns and 1dB. If either parameter (delay or attenuation) is away from its optimum setting, the ‘dip’ for the other parameter is both broader and shallower. Interestingly, at 1dB either side of the optimum attenuation setting the delay settings changed by 1ns. At 0dB attenuation the maximum reduction (29dB) was at 71ns and at 2dB attenuation was at 73ns (24dB). Therefore, even on the bench with a single noise source there was some interaction between the delay and attenuation settings.
The on-air tests were carried out with the three modules, two antennas and cable lengths as in Fig. 1. In this design there is minimal interaction between the delay and attenuator settings so a null can easily be found by first noting the noise level with each antenna separately, putting the attenuator in the path of highest noise, and setting it to give equal noise levels at the receiver. Next, with both antennas connected and the receiver connected to the combiner output, start with the delay at mid-value and use successive approximation to find the optimum. Alternatively, you could, with patience and steady signals and noise, map out the full response as shown below.
Three example audio clips from 40m are available online as a single .wav file at:
Except for the SSB audio clip, where the receiver was a Drake R4B, the receiver for the following tests was the homebrew WSPR direct conversion design with no AGC that I described in April 2016 PW. The three clips are:
- OE6ETF on SSB.
- DL4YDU on CW calling CQ.
- A WSPR station at an audio tone of 1430Hz and DJ6UX calling CQ at 330Hz. This clip starts with no noise cancellation and steps through the spoken delay settings.
For a detailed assessment of performance, I recorded the audio level from the WSPR receiver during a period of steady signals with little fading and measured using the Audacity application on an Apple Mac (it’s also available for Windows). Fig. 9 is the on-air equivalent to Fig. 8, where the noise was measured in a 100Hz band when there were no signals present. Clearly the maximum noise reduction at 12dB was much less than the 36dB achieved on the bench. In addition, the delay and attenuator settings were less critical. Maximum noise reduction was at a delay of 63ns and an attenuator setting of 0dB.
My intuition is that these observations are compatible with multiple noise sources in and around the house, all on the same side of the noise antenna, spanning some 60˚ from a telephone line one side of the house to a street lamp the other. Having reduced the received level of these noise sources, noise from other directions remains. While not spectacular, a 12dB noise reduction is useful and this was on top of a 6dB noise reduction achieved by having the antenna at the far end of the garden away from the house.
This phasing method of noise cancellation also alters the far-field beam pattern of the main antenna. That alteration will depend on the attenuator and delay settings. The result will be alterations to signal levels as well as noise. An early-morning WSPR signal from ON7KB that was both strong and free from QSB provided a good example. Recalling that in these measurements the noise was measured in a 100Hz bandwidth, ON7KB’s signal-to-noise ratio (SNR) was 34dB with no noise reduction and 43dB with optimum settings (delay of 53ns in this case), a 9dB improvement in SNR. The variation in signal and noise levels with delay settings are shown in Fig. 10.
The signal at about 1240Hz is from a QRSS sender. The signal was barely above the noise with no delay (green trace), with the best SNR of about 14dB at a delay of 53ns.
Noise cancellation prior to the antenna socket of the receiver certainly works for me. The simple approach using three passive modules, a switched attenuator, a switched delay line and a combiner, proved effective on SSB, CW and digital modes. However, the noise reduction that can be achieved is likely to be very dependent on the local noise environment, including the spatial distribution of the noise sources and their levels. If you are literally surrounded by multiple noise sources of much the same level, this technique is unlikely to prove effective. Conversely, if you suffer from a single dominant noise source that spans a small angular range at the antennas, your results may approach the bench results of a single source noise reduction of 30dB or more. The online comments on the commercial devices span this range of experiences. They include many tips for getting the most out of this method. If you try this passive modular approach and it does not work for you, at least you will have built three items with a myriad of other uses and probably learnt a bit more about your local noise.
This article was featured in the September 2018 issue of Practical Wireless