VK5HW AR musings: Digital Pre-Distortion (Puresignal)

Linearity

The linearity of an amplifier states that the output of such amplifier increases linearly with the increase of the input signal. Which means that if we have an amplifier which has a gain of 10 dB and if we put 1W into the amplifier we should see an output of 10W. Unfortunately, our linear-amplifiers are not that perfect. Almost all amplifiers behave as depicted below. E.g., they do not follow the ideal.

Due to this nonlinearity our so called linear-amplifiers do create some unsavoury side effects known as Intermodulation Distortion (IMD). Most of us know this as "splatter" !

We can see this phenomenon on signals where either the poor linear-amplifier is being abused, the exciter is faulty or mis-aligned/configured and/or the IMD products for the exiter is less than 30 dB down on the fundamental, a poor exiter. (Side note: There are way's to improve on this!)

With a good panadapter/spectrum display we are able to see and not only hear the IMD, which I might add does contribute to the pollution of our bands. It not only raises the noise level but it also interferes with other amateurs who are trying to use the shared spectrum.

NOTE: Some IMD is less visible on a standard panadapter display because the IMD is hidden IN‐BAND, e.g., it falls within the bandwidth of the stations transmitted signal. And "Yes" this does add unwanted distortion to the transmitted signal !

Pre-Distortion

IMD is an unwelcome and an undesired by-product of any amplifier. But fortunately, for our RF-amplifiers there is this method to reduce IMD using a mechanism called Pre-Distortion. It basically works by making our rf-amplifiers more linear by pre-distorting the input signal. Pre-distorting the input signal in such a way as to offset the distortion at the output of the amplifier. Basically this means that we are able to correct the output to behave very much like an ideal (close to anyway) amplifier.

Please note, that in addition to the amplitude distortion (Pre-Distortion) shown in the below pictures, there is also phase distortion (the amplifier phase shift varies as a function of signal amplitude) which also must be corrected to achieve significant reductions in IMD.

no PD applied

PD applied

Todays compute power of Software Defined Radios (SDR) can easily be used to calculate the required corrections required for PD. These corrections are then applied to the digital transmit samples to create a near perfect output signal.

Please note that Digital Pre-Distortion is also known as "Puresignal" (PS), as implemented by Dr. Warren Platt in the WDSP library. PS continuously monitors and corrects the output signal of the radio or the rf-amplifier. To be able to do so, the Software implementation requires a second digital receiver and, if a rf-amplifier is going to be used in the PS loop, a suitable sensor/sampler (TAP) which will feed a low power signal into the second digital receiver for further processing in the software. This method is called Adaptive Pre-Distortion, as it adapts continuously to the changes in the output signal.

The above is a simplification of what is going on in the PS loop. The correction is actually performed in the Software (WDSP) and the loop is run in the digital domain of the SDR.

Do's and Don'ts

As mentioned above if we use an external amplifier and want to use PS we need a sample of the signal after the amplifier e.g., the signal which is going out towards the antenna. Depending on the sampler/tap used, additional attenuation might be needed to be inserted into the feedback to make sure that the receiver is not going to add additional distortion products to the signal and, of course to protect the sensitive front end of that second receiver.

50dB Sampler/Tap

NOTE: MAKE SURE THAT YOUR FEEDBACK IS ALWAYS SUFFICIENTLY LOW (ATTENUATED) SO IT DOES NOT DAMAGE YOUR HARDWARE.

It is very important that the feedback level does NOT create an ADC Overload situation.
However, to achieve the best PS results, the feedback level must be as CLOSE as practical to ADC Overload.
On my ANAN 100D ADC overload occurs at approximately ‐11dBm (at about 180 mVpp).
Again I can't stress it enough DO NOT CREATE an ADC overload condition.

PA Memory Effects

There is this phenomena in RF-amplifiers called Memory Effects. I try to explain this as best as I possible can. What this means is that the amplifier gain and phase are not only a function of the current input signal but also a function of past input signal(s), e.g. the amplifier remembers the past. These effects can be temperature and/or bias (PWRS) related. This is most prominent using SSB. The signal into the amplifier changes with speech pattern of the operator and so we have strong signals and weak signals into the amplifier. Suppose that a strong signal sends the active device(s) in the amplifier into a hot situation thereby briefly changing its gain and other characteristics. The amplifier will remember this state/characteristics until it cools sufficiently down again. However, in the meantime a number of weaker signals arrive at the amplifier however, they have been predistored for a different state of the amplifier. The end result is PS issues. This is also true for bias and the supply voltages. Which can also change the gain and other characteristics of the amplifier.

Having used PS with a 13.8V 400W - 8dB gain linear-amplifier with voltage supplied from a 60A PWRS, I've detected some memory effects running the amplifier at about 380W output. Dropping the input by 25% solved that issue. There was no issue with the supply voltage, so I assume the issue was more related to heating of the amplifier.

However, using amplifiers with high voltage power FET(s) or LDMOS devices I have not detected any memory effects (YMMV).

So a few points to take away from this to minimise memory effects:

Make sure to monitor the temperature of the amplifier. Using amplifiers with more headroom to keep the amplifier running cool or adding additional cooling should keep memory affects away.
It goes without saying that enough headroom in the PWRS is also required to avoid voltage sag.
Sufficient heat dissipation of all parts.

Puresignal Correction Bandwidth

Another point to consider when operating PS is, PS correction bandwidth. For PS to be able to correctly compute correction, all IMD products need to be within the bandwidth of the receiver used to receive the feedback of the amplifier. PS can only correct what it can "see" within the bandwidth of the channel. The sample rate dictates the bandwidth of the receiver and since this is set in the firmware to be 48ks, the correction bandwidth would be 48kHz. However, we would need to consider a bit of filter roll‐off and the correction bandwidth would be more like 40 kHz.

Eg., +/-20kHz from the center frequency.

For most of us that is not an issue however, for some this could become an issue. What it means is that IMD products need to fall within this 40 kHz bandwidth. PS will NOT work well with very dirty signals e.g., signals which are dirty by overdriving the available hardware.

APPENDIX

The current fav is to use a Hermes Lite, which has about 5W pep output, then add a low voltage intermediate amplifier, before finally driving a tube or a low gain high-power solid state amplifier. If you managed to read all of the above you would have garnered a good understanding now of why that might be fraught with problems. An easier solution would be to forget about the intermediate amplifier and to use a high gain solid state amplifier.