It's interesting.
For the reconstruction filter, a non-constant impedance (ie speaker) changes the last pole and progressive ingress of the effect into each stage of the filter. So for a speaker this would be say between 4-1Kohm impedance range.
For digital RF you could standardise on 50 or 75ohm. The 'output transformer' would then simply be a 50/75ohm primary if possible with the digital switching and . I'm making the assumption that a step down to increase current would then drop the secondary output to say a low 0.75 for example. I would need to check if that could be possible.
MarcelG's reconstruction filter is a LC butterworth that has a stop at 81KHz.
Now the interesting question here is with a high order LC butterworth, how much of a difference does the speaker impedance range make?
This is interesting: Influence of I/O impedance in Butterworth passive filter design - Electrical Engineering Stack Exchange
It suggests that given the impact affects the poles in reverse thus a BW would vary it's upper stop band and response slightly, but it appears not as much as you'd think when you take into account the 81Khz stop band vs say 30-40KHz of the speaker maximum where the pass band still remains relatively unaltered.
There was an interesting paper on IEEE about using stop band to reflect/recycle the RF images/harmonics for power efficiency. However the point about adding a second filer post is an option with the effect on power loss to the speakers.
Lastly the other alternative is to add a zOTL style stage to impedance match and present a constant impedance load to the reconstruction filter. However getting devices running at that speed (IIRC about 10x) may be a thesis just by itself along with the switching harmonics that then need filtering.
For the reconstruction filter, a non-constant impedance (ie speaker) changes the last pole and progressive ingress of the effect into each stage of the filter. So for a speaker this would be say between 4-1Kohm impedance range.
For digital RF you could standardise on 50 or 75ohm. The 'output transformer' would then simply be a 50/75ohm primary if possible with the digital switching and . I'm making the assumption that a step down to increase current would then drop the secondary output to say a low 0.75 for example. I would need to check if that could be possible.
MarcelG's reconstruction filter is a LC butterworth that has a stop at 81KHz.
Now the interesting question here is with a high order LC butterworth, how much of a difference does the speaker impedance range make?
This is interesting: Influence of I/O impedance in Butterworth passive filter design - Electrical Engineering Stack Exchange
It suggests that given the impact affects the poles in reverse thus a BW would vary it's upper stop band and response slightly, but it appears not as much as you'd think when you take into account the 81Khz stop band vs say 30-40KHz of the speaker maximum where the pass band still remains relatively unaltered.
There was an interesting paper on IEEE about using stop band to reflect/recycle the RF images/harmonics for power efficiency. However the point about adding a second filer post is an option with the effect on power loss to the speakers.
Lastly the other alternative is to add a zOTL style stage to impedance match and present a constant impedance load to the reconstruction filter. However getting devices running at that speed (IIRC about 10x) may be a thesis just by itself along with the switching harmonics that then need filtering.
Just to perhaps close off the conversation - I modelled varying speaker impedances vs frequency range in ltspice using MarcelG's 81Khz reconstruction filter used in his valve doc.
Now this doesn't mean much as this is simply an AC sweep, what is important is running a DDS digital signal through vs sweeps of impedance to show the corresponding response.
The filter is designed to remove the aliasing from the switching which it would still do.
However the AC sweep gives a hint at the resulting problems that occur - such as low ohm load results in roll off. With a speaker having a wide range and full range typically an impedance spike around 50-100hz this may result in a uneven frequency response.
Now this doesn't mean much as this is simply an AC sweep, what is important is running a DDS digital signal through vs sweeps of impedance to show the corresponding response.
The filter is designed to remove the aliasing from the switching which it would still do.
However the AC sweep gives a hint at the resulting problems that occur - such as low ohm load results in roll off. With a speaker having a wide range and full range typically an impedance spike around 50-100hz this may result in a uneven frequency response.
Most of the people here understand that the actual impedance of a speaker varies with frequency, and any decent speaker will come with an impedance VS frequency plot.
These plots, however are made using a single frequency sine wave swept slowly over time at a reasonably low power level. They do vary with power level, especially as one or more drivers experience significant excursion.
Even so, we do not listen to single frequency sine waves very often, we tend to listen to music. Sometimes very DYNAMIC music. What do you think happens to the impedance of a speaker, especially a large woofer, when it's cone is moving in one direction at a significant velocity, and a BIG transient comes along and tries to instantaneously reverse its travel?
I set out to answer this question about 20 years ago when 16 bit @ 44 KHz PC recording equipment became affordable, and we had a "music room" where my daughter and her friends annoyed the neighbors. I made recordings of all her drums, the bass guitar, and some big fat synthesizer sounds and mixed them into some speaker / amp torture tests.
I was convinced that a sharp rim shot (drum stick hitting the metal rim of a snare drum) placed in the sine wave of a bass guitar at about 50 Hz could send the instantaneous impedance of an 8 ohm 15 inch pro audio woofer to near zero ohms, possibly even negative, with the speaker stuffing more counter EMF back into the amp than it is eating, especially if the rim shot is placed near the zero crossing of the sine wave.
The further down this rabbit hole I went, the more confused I got, so I eventually assumed that a speaker could do anything it wanted, and every one was different. I also learned a few crazy things about GNFB, and that you should NEVER place a Shure SM-57 microphone INSIDE the bass drum.
Here is a simple experiment that will yield different results from amp to amp. Take a stereo amp and drive one channel with your favorite torture test, even a single frequency sine wave. Short the input of the other channel, load both outputs with a resistive load, then tie another resistor from the hot side of the two outputs together, feeding the output of the "tortured" side into the output of the "quiet" side.
In an ideal amplifier the quiet side should remain quiet if its "ideal" output impedance is zero. No amplifier is ideal, so there will be some residual signal across the output of the quiet channel. It should be a perfect copy of the tortured channel, but at a much reduced level. It should NOT contain additional harmonic content that is not found in the output of the tortured channel, but often does.....
WARNING, this is another DEEP rabbit hole with all sorts of diverging sub tunnels that venture close to the "all GNFB is evil" rabbit hole.
I got out my shovel, covered up all traces of the holes, and haven't looked back in 15 years. I still beat all my DIY amp designs with all sorts of torture including my lousy guitar and bass playing to look for instabilities, and I use as little GNFB as I can, often zero.
These plots, however are made using a single frequency sine wave swept slowly over time at a reasonably low power level. They do vary with power level, especially as one or more drivers experience significant excursion.
Even so, we do not listen to single frequency sine waves very often, we tend to listen to music. Sometimes very DYNAMIC music. What do you think happens to the impedance of a speaker, especially a large woofer, when it's cone is moving in one direction at a significant velocity, and a BIG transient comes along and tries to instantaneously reverse its travel?
I set out to answer this question about 20 years ago when 16 bit @ 44 KHz PC recording equipment became affordable, and we had a "music room" where my daughter and her friends annoyed the neighbors. I made recordings of all her drums, the bass guitar, and some big fat synthesizer sounds and mixed them into some speaker / amp torture tests.
I was convinced that a sharp rim shot (drum stick hitting the metal rim of a snare drum) placed in the sine wave of a bass guitar at about 50 Hz could send the instantaneous impedance of an 8 ohm 15 inch pro audio woofer to near zero ohms, possibly even negative, with the speaker stuffing more counter EMF back into the amp than it is eating, especially if the rim shot is placed near the zero crossing of the sine wave.
The further down this rabbit hole I went, the more confused I got, so I eventually assumed that a speaker could do anything it wanted, and every one was different. I also learned a few crazy things about GNFB, and that you should NEVER place a Shure SM-57 microphone INSIDE the bass drum.
Here is a simple experiment that will yield different results from amp to amp. Take a stereo amp and drive one channel with your favorite torture test, even a single frequency sine wave. Short the input of the other channel, load both outputs with a resistive load, then tie another resistor from the hot side of the two outputs together, feeding the output of the "tortured" side into the output of the "quiet" side.
In an ideal amplifier the quiet side should remain quiet if its "ideal" output impedance is zero. No amplifier is ideal, so there will be some residual signal across the output of the quiet channel. It should be a perfect copy of the tortured channel, but at a much reduced level. It should NOT contain additional harmonic content that is not found in the output of the tortured channel, but often does.....
WARNING, this is another DEEP rabbit hole with all sorts of diverging sub tunnels that venture close to the "all GNFB is evil" rabbit hole.
I got out my shovel, covered up all traces of the holes, and haven't looked back in 15 years. I still beat all my DIY amp designs with all sorts of torture including my lousy guitar and bass playing to look for instabilities, and I use as little GNFB as I can, often zero.
There is an iron clad test to determine how much back EMF an amplifier can eat. Connect the amp in BRIDGE mode, then drive ONE channel. Crank until there is distortion. The driven channel will need to sink the same current the driven channel is sourcing. The back EMF itself isn’t intolerable, but the required current (in the wrong direction) may cause limiting or at least loss of feedback. Power amplifiers, like power HP power supplies, can source much more current than they can sink. Any power amp will do several hundred mA. Typical pro PA amps will do about 2 amps peak per output transistor pair - what the typical short circuit current is intentionally limited to (see the MJ15024 SOA chart as to why). This is all they can do safely, even if the current limit were set higher. Cheap HT amps will go up in flames (so will a Flame Linear). Distortion may set in first before limiting or death occurs. Too much of that reverse current at high frequency can also make the amp unstable. When either side starts oscillating, bad things will happen.
A lot of RF engineers don’t understand this about lab power supplies and it can cause all manner of grief. I ended up building a bootleg supply for the lab that would do +/-12 volts at +/- 2 amps, all four quadrants, without going unstable and used it for about 5 years to bias the gates of depletion mode FET amplifiers. Everybody wanted to use it of course, but the problem was you couldn’t control it with Labview. It had front panel knobs, but that was it. Some techs still refused to use it because they couldn’t code up the bias procedure, so they had to live with explaining away jumps in bias and gain, or compression curves that folded back, or why things were still exploding. Yeah, you CAN put a load on the supply in parallel and fix this, but the measured current reading is way off defeating the purpose of coding it to read the internal meters. They eventually sprung for some very expensive Keithley supplies to get this capability. This IS a back EMF problem - RF energy is rectified to DC (in general, baseband) and sent back into the gate bias supply. Loudspeakers won’t do frequency conversion when doing this, other than making harmonics or IM products. But those are audible.
But why should you NEVER place an SM57 inside of a kick drum? I’m sure it can generate all manner of nasty signal, but when I’m testing a PA amp will will get turned up until you can’t even tell what song is playing anymore. Will the mic itself actually self-destruct? Perhaps generate enough voltage to destroy a 5532 op amp in your mixer board? Do tell, I am curious. I used to make electronic drum pickups by installing 6x9 or 8” speaker drivers inside drums.
A lot of RF engineers don’t understand this about lab power supplies and it can cause all manner of grief. I ended up building a bootleg supply for the lab that would do +/-12 volts at +/- 2 amps, all four quadrants, without going unstable and used it for about 5 years to bias the gates of depletion mode FET amplifiers. Everybody wanted to use it of course, but the problem was you couldn’t control it with Labview. It had front panel knobs, but that was it. Some techs still refused to use it because they couldn’t code up the bias procedure, so they had to live with explaining away jumps in bias and gain, or compression curves that folded back, or why things were still exploding. Yeah, you CAN put a load on the supply in parallel and fix this, but the measured current reading is way off defeating the purpose of coding it to read the internal meters. They eventually sprung for some very expensive Keithley supplies to get this capability. This IS a back EMF problem - RF energy is rectified to DC (in general, baseband) and sent back into the gate bias supply. Loudspeakers won’t do frequency conversion when doing this, other than making harmonics or IM products. But those are audible.
But why should you NEVER place an SM57 inside of a kick drum? I’m sure it can generate all manner of nasty signal, but when I’m testing a PA amp will will get turned up until you can’t even tell what song is playing anymore. Will the mic itself actually self-destruct? Perhaps generate enough voltage to destroy a 5532 op amp in your mixer board? Do tell, I am curious. I used to make electronic drum pickups by installing 6x9 or 8” speaker drivers inside drums.
There was an interesting IEEE paper on recycling reflected RF power from pass band and stop band filters to increase power efficiency. Unfortunately you needed a subscription to read more.
I assume that the microphone in a container with a large membrane acts like a large microphone.. thus low frequency feedback (total system 'global' feedback).. probably ruptured the microphone membrane?
Last edited:
It is normal practice to reflect harmonic energy to increase efficiency. Highest efficiency is when the voltage and current waveforms each contain different harmonics, so that the VI* (power lost) at each of the harmonics approaches zero.
All of my attempts to measure the dynamic impedance of a loudspeaker used tube amps which could eat plenty of reverse power.
It did not survive. After it's death I took it apart expecting to see a smashed diaphragm, but it looked OK. The coil however was open. That mic had been abused for a few years, so it may have been partially damaged before my experiment.
Harmonic energy is normally wasted in the output filter needed to be compliant with spurious emission specs. Harmonic energy recapture was being looked at by other members of my group as far back as the 1990's as were Doherty amps, active load line modulation, Direct RF synthesis and a few others.
I worked in a high efficiency RF power amp design group for several years. Most of my focus was on linear PA's for LTE and other quasi AM modulation schemes that had a high peak to average ratio in the 6 to 12 dB range.
I spent a couple years designing the first working prototype for a power supply modulation system that took real time measurement of the energy in the applied modulation at any given moment, and adjusted the supply voltage on final RF device (usually one or more LDMOS fets) to keep the device just above saturation. Efficiency gains of 10% are possible. This works somewhat like a class H audio amp.
I took this concept and applied it to a vacuum tube amplifier for a magazine design contest back in 2007 where it won a prize and got published. I never did much with it after the article, but someday.....
The article is here:
It did not survive. After it's death I took it apart expecting to see a smashed diaphragm, but it looked OK. The coil however was open. That mic had been abused for a few years, so it may have been partially damaged before my experiment.
Harmonic energy is normally wasted in the output filter needed to be compliant with spurious emission specs. Harmonic energy recapture was being looked at by other members of my group as far back as the 1990's as were Doherty amps, active load line modulation, Direct RF synthesis and a few others.
I worked in a high efficiency RF power amp design group for several years. Most of my focus was on linear PA's for LTE and other quasi AM modulation schemes that had a high peak to average ratio in the 6 to 12 dB range.
I spent a couple years designing the first working prototype for a power supply modulation system that took real time measurement of the energy in the applied modulation at any given moment, and adjusted the supply voltage on final RF device (usually one or more LDMOS fets) to keep the device just above saturation. Efficiency gains of 10% are possible. This works somewhat like a class H audio amp.
I took this concept and applied it to a vacuum tube amplifier for a magazine design contest back in 2007 where it won a prize and got published. I never did much with it after the article, but someday.....
The article is here:
Interesting that you highlight keeping out of saturation. I can’t remember but I think it was a Chord technical talk suggested using class A to keep out of saturation due to the non-linearity.
You cannot operate an RF power amplifier in saturation with an amplitude or phase modulated signal. You need headroom (peak to average ratio) for the modulation or information is lost. An RF amp in saturation puts out a single constant tone signal, at full power. For amplitude modulation you can see why that wont work. But even if all you do is change the phase of the carrier, you get amplitude changes as well. Blame Hilbert. The instant you band-limit (filler) a PM signal you get some AM associated with it. Some schemes have very little amplitude modulation, as little as 2 or 3 dB. Others have a much larger peak to average to cram as much bit rate through a given amount of spectrum as they can. The envelope tracking George describes works because the amplitude only changes at a baseband rate, and you can PWM the power supply fast enough to keep up. Give it just enough voltage to keep it from clipping in real time.
Exactly. This scheme works great at audio, even with vacuum tube voltage levels. My work was in about 2006, used a 200 KHz wide RF channel and a Motorola exclusive modulation scheme called SAM, Scalable Advanced Modulation. It was capable of sending full motion (30 FPS) 640 X 480 video over a 200 KHz channel with minimal compression, but got blown out of the water by WIMAX and then LTE.
I needed to run a 2 MHz sampling and buck converter rate to get the fidelity needed to transmit SAM without degrading the EVM by more than a percent or so. A 10 MHz LTE channel would be a challenge even today.
Nobody runs an RF amp in class A. Even the LTE stuff that requires very good linearity ran class AB, and used a pair of 60 watt LDMOS fets to make 75 watts peak at 793 MHz. This is about 8 to 10 watts average before the duplexer and maybe 4 or 5 watts at the antenna. Linearity came from pre-distortion in the DSP, or Cartesian Feedback, either at baseband, or through the DSP.
I needed to run a 2 MHz sampling and buck converter rate to get the fidelity needed to transmit SAM without degrading the EVM by more than a percent or so. A 10 MHz LTE channel would be a challenge even today.
Nobody runs an RF amp in class A. Even the LTE stuff that requires very good linearity ran class AB, and used a pair of 60 watt LDMOS fets to make 75 watts peak at 793 MHz. This is about 8 to 10 watts average before the duplexer and maybe 4 or 5 watts at the antenna. Linearity came from pre-distortion in the DSP, or Cartesian Feedback, either at baseband, or through the DSP.
You will find some RF PA’s that operate class A. Almost all of them wide band, from octave to decade wide bands. Try to run them in class AB and you can end up with a doubler or tripler. That can happen when harmonics are in band, Especially when you try to add gain slope between stages that tries to prop up the high end. You end up with about 15% efficiency at the upper end of the band, more down at the low end. Can’t use that for no cell phone, you’d get under an hour on a charge.
There are a few wideband class A amps, but there are more class AB distributed amps.
I often used this one for a driver. It does from 100 MHz to 1 GHz at 5 to 10 watts.
I often used this one for a driver. It does from 100 MHz to 1 GHz at 5 to 10 watts.
Note the doubling notch at 425 MHz in Fig, 4. If you want to get rid of that you run it at a higher IdQ. If you’re running over narrow instantaneous BW often you can live with it. But not always. Sometimes they want pancake flat or monotonically decreasing over frequency.
Nick you could ask your question on:
main@funwithtubes.groups.io | Home
I am don't have expertise in TX chains.
main@funwithtubes.groups.io | Home
I am don't have expertise in TX chains.
So, at the moment I can see the main issue is:
Impedance range from the speaker -- influences --> impedance range of the reconstruction filter presented to the RF input side -- causes --> RF reflections and power loss, dynamic spectrum of standing waves if not quelled -- causes --> noise, power loss, high voltage peaks from constructive interference and possibly physical damage.
This is before we get to harmonics, and images.
The impedance differences within the reconstruction filter itself would also cause a microcosm of these same effects between each stage and component.
Simple voltage/impedance bridges don't work with RF. So having a 0.75ohm output and a speaker of 0-1Kohm that is dynamic with both frequency but also musical transient state (as George has described) results in one heck of a headache due to a completely non-static/linear wave model (reflected, standing and passed).
It's possible to create a wave guide to separate reflected waves but the end would then have to have the same impedance at that specific point in time. Almost like both ends of the speaker would have to be connected -- but then you don't want the wave fronts to suddenly cause a constructive peak sending the voltage over the coil insulation. However we're separating RF and speaker in the reconstruction filter.. not the speaker coil (that would be a transmitter as George has stated to previously).
This means either:
a) a very complex RF reconstruction filter that minimises reflection by using wave guides and a curved injection point, thus reflection goes back the other way. between each component..
This would also have to cope wide band given the very dynamic impedance changes. This would as wgski and george has pointed out be exceptionally inefficient, on top of the inefficiency of the amp.. I suspect a very large 150-200W amp driving resulting in 10-20W of musical output. Not to mention the output of the filter in terms of output impedance vs speaker for the audio would cause a rolloff..
b) I standardise and present a constant-ish impedance to one side for example using something akin to the ZOTL within the reconstruction filter, allowing the output to then matched dynamically.
That would then have even more complex fun..
The benefit of running a line level reconstruction filter (ie MarcelG's valvedac) and then have the next stage an impedance matched output for a specific tube as a standard AF amplifier then starts to have a lot of merit in terms of simplicity, lack of reflected wave noise, standing wave issues and other such wonders.
I also just wanted to add that I understand the power envelope for efficiency. I was thinking of running this off the DC link of a SMPS for example initially. Essentially the power supply is the final stage amplifier (sounds very TacT like). The issue is simply if I make the envelope track the peaks and don't have a reserve anyway then we're back to sagging under transients etc. My main concern is more about how much would be lost in the output stage due to impedance.
Impedance range from the speaker -- influences --> impedance range of the reconstruction filter presented to the RF input side -- causes --> RF reflections and power loss, dynamic spectrum of standing waves if not quelled -- causes --> noise, power loss, high voltage peaks from constructive interference and possibly physical damage.
This is before we get to harmonics, and images.
The impedance differences within the reconstruction filter itself would also cause a microcosm of these same effects between each stage and component.
Simple voltage/impedance bridges don't work with RF. So having a 0.75ohm output and a speaker of 0-1Kohm that is dynamic with both frequency but also musical transient state (as George has described) results in one heck of a headache due to a completely non-static/linear wave model (reflected, standing and passed).
It's possible to create a wave guide to separate reflected waves but the end would then have to have the same impedance at that specific point in time. Almost like both ends of the speaker would have to be connected -- but then you don't want the wave fronts to suddenly cause a constructive peak sending the voltage over the coil insulation. However we're separating RF and speaker in the reconstruction filter.. not the speaker coil (that would be a transmitter as George has stated to previously).
This means either:
a) a very complex RF reconstruction filter that minimises reflection by using wave guides and a curved injection point, thus reflection goes back the other way. between each component..
This would also have to cope wide band given the very dynamic impedance changes. This would as wgski and george has pointed out be exceptionally inefficient, on top of the inefficiency of the amp.. I suspect a very large 150-200W amp driving resulting in 10-20W of musical output. Not to mention the output of the filter in terms of output impedance vs speaker for the audio would cause a rolloff..
b) I standardise and present a constant-ish impedance to one side for example using something akin to the ZOTL within the reconstruction filter, allowing the output to then matched dynamically.
That would then have even more complex fun..
The benefit of running a line level reconstruction filter (ie MarcelG's valvedac) and then have the next stage an impedance matched output for a specific tube as a standard AF amplifier then starts to have a lot of merit in terms of simplicity, lack of reflected wave noise, standing wave issues and other such wonders.
I also just wanted to add that I understand the power envelope for efficiency. I was thinking of running this off the DC link of a SMPS for example initially. Essentially the power supply is the final stage amplifier (sounds very TacT like). The issue is simply if I make the envelope track the peaks and don't have a reserve anyway then we're back to sagging under transients etc. My main concern is more about how much would be lost in the output stage due to impedance.
Last edited:
Are so you are looking at DSD512 missed this, i.e. 1-bit DAC all the way to the speaker so class D. So I think you would need an active cathode follower a reconstruction filter and a transformer. Sometimes MOSFETS are much better.
Yes, essentially. Although I tend to think of class D being PWM rather than PDM (DSD).
From OTL experience I'm happy to hybridise the output stage. So you could look at Class D output. I'd probably stick with MarcelG's approach of using a balanced filter.
I had enclosed the data sheet in my previous post. I used this part in several designs during my career at Motorola which ended in 2014. The part is NRND now, though DigiKey still has a few in stock.
It reveals a power output of 11 watts @ 100Mhz with an efficiency of 55%, so 20 watts of DC into the device gets you about 11 watts of RF out at 100 MHz. This degrades to 6.5 watts output @ 900 MHz with a 34% efficiency, so 19 watts of DC into the device gets you 6.5 watts of RF out at 900 Mhz.
The part is internally matched to 50 ohms across the 50 MHz to 950 MHz range. It also has a nearly constant 23 dB gain from 100 MHz to 930 MHz. This means that a harmonic filter is required on its output if this will be used in a transmitter application. I used a switched filter bank to cover 136 to 936 MHz. My filter had over 1 dB of loss at 900 MHz considerably reducing the available power to the antenna.
There are probably better choices for a broadband transmitter today, but I really haven't paid much attention to the RF device world since I left Motorola.