Nope, sorry, Tom is correct in this.

Splatter is an effect that occurs close in to the transmitted frequency and is due to intermod between components of the signal being amplified.  A good solid state amp running class AB might have low order 2 tone IMD products at around the -36dBc level, but these fall off fairly quickly as you move outside the nominal bandwidth so are usually more of an audio quality problem then a source of interference as long as the amp is not over driven.
The really nasty splatter that results from overdrive is really RF clipping as the stage runs out of supply voltage to allow it to track the envelope, this will also cause an increase in generated harmonics (as the stage becomes very non linear), but the mess in band will be a bigger problem.

The harmonic generation tends to be due to imbalance between the two phases of the circuit (even harmonics) and the same non linearities that cause splatter (odd harmonics), but this time the mixing products are typically products of signals mixing with themselves rather then 2 tone products.

A push pull stage helps with lowering the even harmonics, but will typically not be good enough on its own to meet spec, and the output low pass will need to go over somewhere below twice the operating frequency to add additional suppression to the second harmonic output, for all that the third harmonic needs more suppression then the second.  

These filters are only peripherally concerned with splatter in that unless you design the filters as diplexers the harmonic energy gets reflected back into the output of the PA and can itself cause in band IMD2 components! This is however a second order effect and about the only amps that really deal with this tend to be homebrew and sometimes end up in QEX, DUBUS or such.  

Look at it this way, splatter is in the band you are working, harmonics are at multiples of the working frequency, filters help with harmonics but not with splatter. Of course the harmonic levels may track a function of the modulation envelope, but that does not make them splatter.  

73's Dan. M6ATV.

The buck transformer can be smaller then you think as it does NOT have to handle 1.5KW.

If you want to buck say 10% off a 120V line at 1.5KVA nominal load  you need a buck transformer that can handle 1500/120 = 12.5A @ 12V, which equals only 150VA. Something like a 120V primary to a 12V secondary at 150VA should do the job.

Incidentally that PA circuit does not by any means tell the whole story for a solid state PA, in particular the output filters (Vital) and (all important) protection circuitry is missing, as are the power combiners and splitters, TR switching. Also I have my suspicions about the linearity of that circuit, doing a really linear PA takes rather more work then that!

Regards, Dan.

Sensitivity to mismatch will obviously be higher and I would be careful to consider what happens under high line voltage conditions, but my instinct is that (assuming those are DC output voltages not RMS ones) it should work, albeit possibly with reduced reliability. 

Personally I would fit a small buck transformer to reduce the primary voltage seen my the main iron, RF power devices, particularly of the older sort, do not often have much conservatism built into the manufacturers ratings.

Regards, Dan.

They would also need to buy a commercial license from MIT for fftw (The  fast fourier library used) at a minimum, it is at least available,  and possibly sort out some other dependencies.
Generally taking code with free software dependencies closed source is a pain unless you have planned to be able to do it from the get go.

Regards, Dan.

You probably do not want to use a conventional receive path for the error term as it will almost certainly have too much group delay or too little bandwidth and loop stability will become seriously problematic.   

Cartesian feedback works and can now (with the advent of low cost DDS parts) be made to work for a multi band set without major pain, but it is far more applicable to a integrated transmitter then it is to something designed to work with an add on power amplifier.
The issue is one of loop bandwidth as the group delay through the PA can (as long as it is small compared to the highest modulating frequency), be compensated by tweaking the phase of the local oscillators for the forward and reverse paths.

I have a design I am working on that I may try to get into Radcom or QEX at some point in the future.

There is a paper from IIRC Marconi dated sometime late 70's or early 80's that discusses an approach to correcting the phase errors in the feedback term. 

They quote 2 tone IMD levels of better then -60dBc (As compared to maybe -30 or so for an uncorrected PA).

The place where this becomes really useful is for folks experimenting with advanced digital modes where things like OFDM impose really nasty linearity requirements on the amplifiers.

As for class E, I suspect that the way to skin that cat is to switch the output network as well as the lowpass filter on a per band basis. You are already switching the lowpass network so moving the relay to the drain side of the tank should not be that hard (for all that the voltages are nasty).
A polar loop would probably be required to linearize such a rig, but it is worth thinking on.
It should be possible to design a switchmode modulation supply that could run directly off the mains and would provide direct conversion to the required envelope drive, while the carrier is produced with appropriate phase modulation at low power. All the switching fets and modulation amplifier circuitry would be common to all bands with the output tank and LPF switched per band.

None of this is unique to a SDT or course, but some DSP will make bits of it easier.

73's Dan M6ATV.

Those radios prove that it is possible to build a good radio that way, which is not really in doubt.

My contention is that IF narrowband is all you need then it is possible to build a better radio by specializing the hardware for the purpose then it is by passing everything through to the AD.
The tradeoff of course is that you cannot then do wideband things.

The QS1R has a MDS (in 500Hz bandwidth) of -111dBm on 20M, P3AKEs frontend hits -135dBm under the same conditions, with IMD3DR or 123db.
That 14dB can either be taken as extra sensitivity, or via the attenuators as a higher IIP3 by switching in the attenuator, put that on line and the wideband IIP3 hits something like +65dBm.
At that performance level the LO phase noise becomes an interesting design exercise (and one that is still causing head scratching).

Now the QS1R does all kinds of cool things that a RX with a narrow filter never will, so as I say you need both.

Regards, Dan.

Thats a pretty big if Brian!

Depending on what you mean by wideband, with a 100Mhz ADC you could have DC to ~40Mhz or so, assuming you can handle the data rate (might need a FPGA or something), but that includes everything from big MW and LW Broadcast sites (not to say the 40M broadcast band that can put a mW into a 40M dipole on its own!), right down to some PSK31 from half way around the planet running 5W....

Not to mention 40Mhz worth of atmospheric and thermal noise!

The problem becomes exponentially easier as the rx bandwidth reduces and as you say that 100Mhz ADC could be decimated to add word length at the lower rate, provided its differential nonlinearity is good enough to avoid IP2 or IP3 becoming the limit.

ADC aperture jitter will of course add the the sampling clock phase noise problem, but I dont know how well that has been resolved in modern parts.

I think that from an implementation perspective wideband and weak signal work are pretty directly opposed simply because of the reality of the available components and their linearity. 

I must however have a close look at the specs on something like a flex 5K just to confirm this for a commercial state of the art rig.

Regards, Dan.

The usual limit on the current generation of power MOSFETs seems to be thermal when used in linear service.

Several of the manufacturers have the bad habit of quoting dissipation at 25 degrees case temp, and in some cases when you derate for something you might actually manage to maintain, 75% of the rating goes away!

I would note that the test in that video has a rather short length of line shorting, so it may have been placed at a 'favorable' phase angle, but for all that, more robust parts even if you have to take the robustness with a pinch of salt is something to be welcomed.

I wonder if that VHF device would turn out to be acceptably stable (and reasonably linear) in a broadband HF rig? I am thinking maybe 4-500W output from a 1100W theoretical device would be a more reasonable target from a thermal management perspective.

As to power supplies, EBay, HP server DC supplies, 48V 3000W, probably get one for $50 or so.

Regards, Dan.

So many projects, so little time.!
Going to use the blackgates stuff or build from scratch?

I have a (very) part built simplex in 7.5, but ask me for anything but time!

Regards, Dan.

While we are talking about BDR with respect to signals lying within the IF passband there is on a theoretical level very little difference between a wide and narrow band IF strip, but where an SDR effectively ONLY has the BDR level that applies for signals inside the IF passband, a narrow band rig often gets MUCH better as soon as you can place the interference outside the first IF bandwidth.

A narrow band IF means that it is usually possible to tune such that a strong signal a few hundred Hz away lies outside the IF passband with the narrow filters engaged, which you cannot do with a wideband IF. 
The effect is that in a narrow band RX the performance limitation is usually the first mixer IIP3, while in a wideband design it is likely to be the IF gain stage IIP3, and associated AGC and converter ranges.

The key point is that with a narrow band IF the front end selectivity will allow a weak signal to be worked relatively close to a strong one without both signals falling inside the IF passband and thus without the later stages needing that much dynamic range, a wideband IF does not have that option and must be sufficiently linear to resolve two signals that may be 100dB or so different in strength when both fit inside the IF passband.

ADCs are interesting beasts as it seems to be that the product of bandwidth and word length is largely (for any given generation of parts) constant, but the issue is only partly the ADC, it is also the gain stage that in all probability precedes it.  I would note that there are few (if any) high speed 24 bit converters that really manage a thermal noise level better then 20 or so bits below full scale, so the dithering is unlikely to be an issue.

It is true that from an information theory perspective I am throwing away bits, but they are bits which (as far as my wanted signal is concerned) contain only noise anyway, so why do I care, if wideband is your thing then a different tradeoff may be appropriate.


Regards, Dan.

I suspect that we may see a range of SDRness in the market, starting at one end with direct digitization right at the aerial and proceeding all the way to no DSP at all at the other extreme.
I could certainly see  several points in a typical RX where you could (with varying engineering tradeoffs) decide to put the converter.

Question: Does SDR necessarily imply wideband input to the ADC, or is an equally valid implementation to only digitize after cutting the IF bandwidth way down? After the IF amplifier/AGC? Actually at baseband?
Each of these trades off some possible software capabilities for (probably) better BDR.

Now I have to admit to having PA3AKEs awesome receiver board and quartz set on order but I am giving serious thought to going DSP for the detector or possibly doing the DSP at baseband, in any case it will be after some brutal IF filtering, so I probably cannot get much of a waterfall out of it as the IF may have been butchered down to a mere 250Hz wide by the time the DSP gets a look.

Does a DSP based demodulator behind a conventional front end still count as SDR?

Just some thoughts to ponder.



Regards, Dan.

That is fine for a rig sold expressly as a SDR,with all the compromises for conventional narrow band working that implies, but it will not compete with a conventional radio for digging weak CW out of a massive pileup on say 160M because the dynamic range will not be there. That few hundred Hz wide first IF filter in a conventional rig is there for a reason. You really need both to be available, SDR for doing what it does and a conventional narrow IF set for when that is appropriate.
Maybe a SDR 'second' receiver so you have the best of both worlds in one box.

Now personally I despise touchscreens for anything that needs frequent use, for all that things like AGC parameters and other occasionally tweaked things could usefully go there, but from an operator ergonomics perspective I need main tuning/rit/volume/IF Gain/band selection/mode and possibly a few other things on hardware controls I can find without looking, all the infrequently used stuff can go to touchscreen.   

Separating the control surface from the radio is however an obvious direction to move in, a simple serial link between the two should be nearly trivial. 

Regards, Dan.

It seems to be money but no time or time but no money, I too am in the time but no cash phase at the moment, software work is a bit thin on the ground at present and the fact that I never bothered to get a degree does not help.

You do live steam as well (He says meaningfully looking at a well used colchester and tom senior amid the usual pile of swarf)!
Out with it, what loco?

I have to admit to tending to build 'scaleish' things that never actually existed rather then slavishly copying real engines, but it is all good. 

The examiner does not really have to understand it, the UK practical is basically to make sure you can do a workmanlike job of whatever you build, and given that one of the tutors is an ex agilent design engineer I am not to worried.
Sure, I could just build a morse practice oscillator or direct conversion RX set or something of the sort but frankly, boring!

Regards, Dan.

True, but I am already running much lower bias current then is normal for those transistors because I don't need the amp to have spectacular uncorrected linearity. The logic that shuts the power stage down on receive could I suppose be modified to provide for variable bias fairly easily, I might experiment with it.

In fact there is another trick that variable bias would enable: For CW the PA can sit in AB at the start and end of each pulse but once the power has stabilized the bias can ramp down and the drive power up so as to give class C operation. The keying envelope should have a few ms rise and fall time during which the amp would need to be linear but the bulk of the time such a stage would run with zero bias.

I guess I just find the notion of going to class A PA strips as some of the big manufacturers have done, at least to moderate power levels, fundamentally inelegant.

That 7.5KHz wide band limit is not met by a lot of the commercial radios under modulation, especially when hooked up to an external amplifier and somewhat over driven, but is actually not hard to meet this way.

What does go pearshaped is overmodulation as the PA runs out of stick and the loop tries to compensate, splatter for days.... A fast limiter monitoring I,Q, and the required envelope helps solve this.

Interestingly my cartesian loop board is almost all analogue with the exception of two DDS chips used for the forward and reverse path LO and the micro that programs them and monitors a couple of phase comparators. The trick is that both DDS are clocked by the same master clock and so run at exactly the same frequency, however the relative phase can be easily adjusted to control the feedback phase and compensate the delay inherent in the amplifier and output filter.

I am thinking of using this as my required intermediate license project (But probably just the cartesian loop part).
 
Regards, Dan.

Modulating the bias does not however change the mismatched collector load impedance when the amplifier is not producing full output.

Consider a PA designed for say 100W with a 50V rail (and ignore device saturation), the design drain load will be Rl=50^2/200 = 12.5 ohms.
Now consider the same amplifier running at say 10W, with the same load impedance we get a drain voltage swing of V^2 = 2RP=sqrt 250 = 16V, with the other 34V going purely as heat in the transistors.

With SSB modulation having a peak/average ratio of say 4 on a good day, that amounts to a massive heat reduction in any mode except FM, RTTY, some data and CW.

While designing a suitably fast power supply used to be a major issue there are a new generation of switchmode audio amp parts available that offer some possibilities.

My cartesian loop scheme does work and does not appear to increase out of band emissions (careful consideration of loop bandwidth and very careful noise analysis of the error amplifiers and master oscillator, two carefully placed AD797s also help with the noise), at least according to my old Marconi spectrum analyzer.

I cannot of course put this on the air until I upgrade my license, waiting for the (almost trivial) exam now.   
 
Regards, Dan.