One more thought about your problem: you are dumping ~15V of voltage (and power) in the form of heat. A sensible advice would be: first try to reduce that drop, by using another output/winding/supply.
If it is too inconvenient (something I can understand), you can use it to your advantage: you can dump most of it in a resistor, dimensioned to let enough margin for the regulator under worst-case current-draw conditions (a minimal bypass directly in front of the regulator will be required for stability).
Power resistors are designed to dissipate power safely and reliably, and if you have such a current-limiting device, a protection component like a transzorb, transil or similar across the load will protect it until the fuse eventually melts.
If it is too inconvenient (something I can understand), you can use it to your advantage: you can dump most of it in a resistor, dimensioned to let enough margin for the regulator under worst-case current-draw conditions (a minimal bypass directly in front of the regulator will be required for stability).
Power resistors are designed to dissipate power safely and reliably, and if you have such a current-limiting device, a protection component like a transzorb, transil or similar across the load will protect it until the fuse eventually melts.
Alternative frequency synthesis method
A very important method of frequency synthesis is the heterodyning of two signals. It is a very powerful method, and it is widely used in many fields.
The mixing process is performed in a multiplier, and ideally, the mixing products are only Fc+Fm and Fc-Fm (Fc being the "carrier" frequency, and Fm the "modulation" signal; these are arbitrary terms).
The reality is slightly less clear-cut, but let's concentrate on the purely theoretical case: normally, we only need one of the modulation products; the other has to be eliminated in some way.
A commonly used technique is filtering: this is relatively easy if Fc+Fm and Fc-Fm are far apart, and do not vary too much.
When Fm is small, very sharp filters are required, and if one or both of the frequencies are highly variable, the problem is greatly complicated.
A method, based on trigonometric identities can generate directly the wanted product, and suppress the other:
If you generate the two products:
cos(c)*cos(m)=0.5*cos(c-m)+0.5*cos(c+m)
sin(c)*sin(m)=0.5*cos(c-m)-0.5*cos(c+m)
and compute the sum and the difference, one of the terms will be cancelled, and you will be left with a pure (ideally) c+m or c-m.
This method doesn't require filtering, and is sound from a theoretical point of view, but it requires two multipliers (modulators), and two Hilbert function blocks, to generate the quadrature signals, and that is a major issue, because phase-shifting a signal by 90° for a wide frequency range is no mean task.
In addition, the method only works with linear function blocks and pure sinusoidal signals, and all the operations, multiplication, addition/subtraction and Hilbert transformation must be carried out accurately for a good suppression of the unwanted sideband.
It would be nice to be able to use the same kind of trick with purely digital, logic signals, but things don't work like that: using a XOR gate, you can mix two frequencies and generate their sum, but for the average frequency only: the output pulse train will be completely irregular, causing an immense jitter.
Similarly, using a "swallow" circuit, you can subtract two frequencies, but again the output pulse train will have gaps.
In general, digitally combining two random pulse trains results in an awful mess
Thus, with squarewaves only, a clean synthesis is impossible.
The next best waveform is a triangle: it can be generated easily from a squarewave.
Would it be possible to mimic the trigonometric computation trick for triangle and squarewave using an adapted mixer (simple voltage comparator) and logic?
Yes, and here it is:
For clarity, this first pic shows only the various input signals: they consist of the two initial squarewaves to be added or subtracted, their integrated versions (triangles), and the complement of one of the triangle.
Here, only two input triangles are shown, with the final sum output:
It is clear that the yellow trace is the sum of the two initial frequencies, despite having no harmonic relationship with any of them, and it is perfectly regular.
Thus the method works; it is relatively demanding regarding input signals and would be difficult to apply at frequencies of hundreds of MHz, but it is perfectly usable for some applications.
For example, if you want to finely adjust the speed of a crystal-controlled, DD turntable without modifying the circuit or substituting the crystal for a frequency synthesizer, you could intercept the clock signal at a convenient frequency, around 100kHz for example, and insert this circuit. The fixed 100kHz would be converted to a triangle, and a low-frequency function generator would inject the variable tweaking frequency, from ~0 to 500Hz for example.
Nothing very difficult, and all of this can be performed with a good accuracy (the triangle amplitudes must be equal), which will produce a good quality, jitter-free output.
Any residual quirk will easily be polished off in the subsequent frequency dividers chain.
Other similar applications can certainly be considered.
A very important method of frequency synthesis is the heterodyning of two signals. It is a very powerful method, and it is widely used in many fields.
The mixing process is performed in a multiplier, and ideally, the mixing products are only Fc+Fm and Fc-Fm (Fc being the "carrier" frequency, and Fm the "modulation" signal; these are arbitrary terms).
The reality is slightly less clear-cut, but let's concentrate on the purely theoretical case: normally, we only need one of the modulation products; the other has to be eliminated in some way.
A commonly used technique is filtering: this is relatively easy if Fc+Fm and Fc-Fm are far apart, and do not vary too much.
When Fm is small, very sharp filters are required, and if one or both of the frequencies are highly variable, the problem is greatly complicated.
A method, based on trigonometric identities can generate directly the wanted product, and suppress the other:
If you generate the two products:
cos(c)*cos(m)=0.5*cos(c-m)+0.5*cos(c+m)
sin(c)*sin(m)=0.5*cos(c-m)-0.5*cos(c+m)
and compute the sum and the difference, one of the terms will be cancelled, and you will be left with a pure (ideally) c+m or c-m.
This method doesn't require filtering, and is sound from a theoretical point of view, but it requires two multipliers (modulators), and two Hilbert function blocks, to generate the quadrature signals, and that is a major issue, because phase-shifting a signal by 90° for a wide frequency range is no mean task.
In addition, the method only works with linear function blocks and pure sinusoidal signals, and all the operations, multiplication, addition/subtraction and Hilbert transformation must be carried out accurately for a good suppression of the unwanted sideband.
It would be nice to be able to use the same kind of trick with purely digital, logic signals, but things don't work like that: using a XOR gate, you can mix two frequencies and generate their sum, but for the average frequency only: the output pulse train will be completely irregular, causing an immense jitter.
Similarly, using a "swallow" circuit, you can subtract two frequencies, but again the output pulse train will have gaps.
In general, digitally combining two random pulse trains results in an awful mess
Thus, with squarewaves only, a clean synthesis is impossible.
The next best waveform is a triangle: it can be generated easily from a squarewave.
Would it be possible to mimic the trigonometric computation trick for triangle and squarewave using an adapted mixer (simple voltage comparator) and logic?
Yes, and here it is:
For clarity, this first pic shows only the various input signals: they consist of the two initial squarewaves to be added or subtracted, their integrated versions (triangles), and the complement of one of the triangle.
Here, only two input triangles are shown, with the final sum output:
It is clear that the yellow trace is the sum of the two initial frequencies, despite having no harmonic relationship with any of them, and it is perfectly regular.
Thus the method works; it is relatively demanding regarding input signals and would be difficult to apply at frequencies of hundreds of MHz, but it is perfectly usable for some applications.
For example, if you want to finely adjust the speed of a crystal-controlled, DD turntable without modifying the circuit or substituting the crystal for a frequency synthesizer, you could intercept the clock signal at a convenient frequency, around 100kHz for example, and insert this circuit. The fixed 100kHz would be converted to a triangle, and a low-frequency function generator would inject the variable tweaking frequency, from ~0 to 500Hz for example.
Nothing very difficult, and all of this can be performed with a good accuracy (the triangle amplitudes must be equal), which will produce a good quality, jitter-free output.
Any residual quirk will easily be polished off in the subsequent frequency dividers chain.
Other similar applications can certainly be considered.
Thanks, but I can return the compliment, and in addition you have also managed to be or remain more rigorous and thorough than me: the equations I write are inexact and sloppy, have no units consistency, no constants, etc.
Many years ago, I did those things more or less properly, a bit like you, but my standards have declined, and I became sloppier and sloppier with time.
When I really need to do things properly, I am still able to, but it is laborious and demands a lot of time and sweat.
You have been able to retain your orthodoxy AND creativity, which is good, because in technical and scientific fields, exactitude is paramount
Many years ago, I did those things more or less properly, a bit like you, but my standards have declined, and I became sloppier and sloppier with time.
When I really need to do things properly, I am still able to, but it is laborious and demands a lot of time and sweat.
You have been able to retain your orthodoxy AND creativity, which is good, because in technical and scientific fields, exactitude is paramount
I wouldn't be too sure of that. The most impressive things sometimes just net quiet contemplation. Like the finish of an inspired music at concert , everybody in rapt silence until they remember their duty to make noise with their hands.
I have presented an experimental/concept proof amplifier based on the Tandem topology here:
https://www.diyaudio.com/community/threads/tandem-based-amplifiers.388400/post-7074596
https://www.diyaudio.com/community/threads/tandem-based-amplifiers.388400/post-7074596
Now, to lighten the mood, a small interlude on the subject of transmission lines. Nothing especially fundamental or groundbreaking, but entertaining and possibly useful.
First:
A mini-TDR:
This tiny contraption is a multimeter add-on converting the length of a cable into a voltage, with 1mV=1m. With a regular 2kpts multimeter, this results in a resolution of 10cm. With a 20,000pts meter, like in this example, the resolution becomes 1cm (and it remains perfectly usable):
A nice feature is the ability to work automatically with open or shorted cables: any type of impedance discontinuity is identified and usable. The cable cannot be terminated by Zo obviously, and the impedance range needs to be selected, as well as the speed factor, dependent mainly on the dielectric type and conductor spacing for symetrical pairs.
When properly calibrated, it is a useful piece of kit, allowing you to accurately estimate the length of a cut of cable, without unspooling it or worry about its termination.
Most cables on a spool are terminated with an open, but sometimes the cut isn't clean and results in a short. The tester does both, transparently (and economically).
Here is the circuit:
A 50kHz squarewave generated by an oscillator/frequency divider combo is sent into the line to be tested, and the circuit applies a step to a filter whilst no discontinuity is detected. The width of the step relative to the 20µs period creates a PWM signal representative of the line length.
The components used show its age, but more modern ICs than the 74H04 are certainly usable. It could also be built as a stand-alone instrument.
The tester is natively unbalanced, but it works perfectly with balanced cables: the small size of the tester + battery operated multimeter presents a small capacitance wrt. the ambient space, and does not disturb the measurement.
First:
A mini-TDR:
This tiny contraption is a multimeter add-on converting the length of a cable into a voltage, with 1mV=1m. With a regular 2kpts multimeter, this results in a resolution of 10cm. With a 20,000pts meter, like in this example, the resolution becomes 1cm (and it remains perfectly usable):
A nice feature is the ability to work automatically with open or shorted cables: any type of impedance discontinuity is identified and usable. The cable cannot be terminated by Zo obviously, and the impedance range needs to be selected, as well as the speed factor, dependent mainly on the dielectric type and conductor spacing for symetrical pairs.
When properly calibrated, it is a useful piece of kit, allowing you to accurately estimate the length of a cut of cable, without unspooling it or worry about its termination.
Most cables on a spool are terminated with an open, but sometimes the cut isn't clean and results in a short. The tester does both, transparently (and economically).
Here is the circuit:
A 50kHz squarewave generated by an oscillator/frequency divider combo is sent into the line to be tested, and the circuit applies a step to a filter whilst no discontinuity is detected. The width of the step relative to the 20µs period creates a PWM signal representative of the line length.
The components used show its age, but more modern ICs than the 74H04 are certainly usable. It could also be built as a stand-alone instrument.
The tester is natively unbalanced, but it works perfectly with balanced cables: the small size of the tester + battery operated multimeter presents a small capacitance wrt. the ambient space, and does not disturb the measurement.
Another way of detecting impedance discontinuities can be found here: https://www.diyaudio.com/community/...m-beyond-1-the-stufinator.304076/post-4995295
On the theme of transmission lines, here is another small, but entertaining project:
A magical ohmmeter
The specs might look feeble for something "magical": it has a single 200 ohm range, and that's it, but the magic is elsewhere, it resides in its universality: it can measure anything -literally-
The characteristic impedance of a cable for example:
Or a delay line (and many other line-related devices):
It can even measure an ordinary resistor, but you don't need magics to do that: any ordinary ohmmeter will do.
At first sight, it looks impossible: normally such measurements require sophisticated test-instruments, like vector network analyzers, and an access to both ends of the cable. By contrast, Magix only requires an access to one end, with the distant side completely indifferent: it can be open, shorted or terminated by any passive impedance.
Sometimes, when a rookie or an intern arrives for a new job in an electronics-oriented company, a prank played by elders is to task him/her with the quality-control of freshly delivered cables.
The poor guy/gal is sent to the reception area, equipped with an ohmmeter, with the mission of checking the accuracy of the characteristic impedance, and he struggles for half a day with cables spools as high as himself, trying to figure out a way to extract a meaningful reading... which never happens, of course.
However, with Magix, he could easily make the measurement.
How is it possible?
In fact, it is actually an ordinary ohmmeter, sending a calibrated current and recovering a voltage drop. However, to make it work with a transmission line-like device, the measurement has to be completed before the current step has the time to make the round-trip through the DUT.
In principle, with a very (very!) long cable, it could be done manually: with a cable as long as the earth circumference, the round-trip distance is 80,000km, you need to complete the measurement in <~0.4s, which would just be feasible manually... except it wouldn't work: real cables have parasitic parameters, series resistance in particular, and this means that the apparent |resistance| would increase with time, tending to infinity for long times/length.
The trick is to make the measurement very quickly: <20ns for Magix, which translates into ~2m for ordinary cables.
This is how it is done:
A pulse formed by a simplified artificial line is used to create the current stimulus and a sampling window for the voltage measurement (Q1). Q2 is the sampling switch, and J1 resets the line before the measurement.
The voltage held on C1 is sent to a small DPM. It works like a conventional 7106-based one, except all the voltages are ~1/3rd: supply, etc. It was salvaged from a cheap battery tester, and I bought a dozen of them just to cannibalize them. They are tiny and low-power, perfectly suited to small instruments:
Additional corrections are required, because the CCS and sampler are imperfect, due to Early effect, etc. This causes a compression of the scale which is too large to be accommodated with just slope+offset corrections:
The solution is to modulate the general supply voltage by the measured voltage, to straighten the curve: that's the role of R23.
A view of the (gory) innards:
The tester is unbalanced, but as it is battery-operated and electrically small, it can be used on balanced lines too. A common-mode ferrite can be used to perfect the measurement:
The limit of 2m is a bit long, but the circuit is very simplified, and based on 74HC logic. With real lines, and 74AC logic, this version can go down to <1m:
However, this also means that the connection to the DUT needs to be SOTA, because the measurement spectrum extends higher, and is more demanding. The 2m version is more tolerant.
For those interested, more detailed info can be found here: https://forums.futura-sciences.com/projets-electroniques/384792-un-ohmmetre-magique.html#post2889068
On the theme of transmission lines, here is another small, but entertaining project:
A magical ohmmeter
The specs might look feeble for something "magical": it has a single 200 ohm range, and that's it, but the magic is elsewhere, it resides in its universality: it can measure anything -literally-
The characteristic impedance of a cable for example:
Or a delay line (and many other line-related devices):
It can even measure an ordinary resistor, but you don't need magics to do that: any ordinary ohmmeter will do.
At first sight, it looks impossible: normally such measurements require sophisticated test-instruments, like vector network analyzers, and an access to both ends of the cable. By contrast, Magix only requires an access to one end, with the distant side completely indifferent: it can be open, shorted or terminated by any passive impedance.
Sometimes, when a rookie or an intern arrives for a new job in an electronics-oriented company, a prank played by elders is to task him/her with the quality-control of freshly delivered cables.
The poor guy/gal is sent to the reception area, equipped with an ohmmeter, with the mission of checking the accuracy of the characteristic impedance, and he struggles for half a day with cables spools as high as himself, trying to figure out a way to extract a meaningful reading... which never happens, of course.
However, with Magix, he could easily make the measurement.
How is it possible?
In fact, it is actually an ordinary ohmmeter, sending a calibrated current and recovering a voltage drop. However, to make it work with a transmission line-like device, the measurement has to be completed before the current step has the time to make the round-trip through the DUT.
In principle, with a very (very!) long cable, it could be done manually: with a cable as long as the earth circumference, the round-trip distance is 80,000km, you need to complete the measurement in <~0.4s, which would just be feasible manually... except it wouldn't work: real cables have parasitic parameters, series resistance in particular, and this means that the apparent |resistance| would increase with time, tending to infinity for long times/length.
The trick is to make the measurement very quickly: <20ns for Magix, which translates into ~2m for ordinary cables.
This is how it is done:
A pulse formed by a simplified artificial line is used to create the current stimulus and a sampling window for the voltage measurement (Q1). Q2 is the sampling switch, and J1 resets the line before the measurement.
The voltage held on C1 is sent to a small DPM. It works like a conventional 7106-based one, except all the voltages are ~1/3rd: supply, etc. It was salvaged from a cheap battery tester, and I bought a dozen of them just to cannibalize them. They are tiny and low-power, perfectly suited to small instruments:
Additional corrections are required, because the CCS and sampler are imperfect, due to Early effect, etc. This causes a compression of the scale which is too large to be accommodated with just slope+offset corrections:
The solution is to modulate the general supply voltage by the measured voltage, to straighten the curve: that's the role of R23.
A view of the (gory) innards:
The tester is unbalanced, but as it is battery-operated and electrically small, it can be used on balanced lines too. A common-mode ferrite can be used to perfect the measurement:
The limit of 2m is a bit long, but the circuit is very simplified, and based on 74HC logic. With real lines, and 74AC logic, this version can go down to <1m:
However, this also means that the connection to the DUT needs to be SOTA, because the measurement spectrum extends higher, and is more demanding. The 2m version is more tolerant.
For those interested, more detailed info can be found here: https://forums.futura-sciences.com/projets-electroniques/384792-un-ohmmetre-magique.html#post2889068