Wideband (<30MHz) AM modulator circuit, how to make one?

[Yansi]

Hello,
my curious me again, investigating topics I probably shouldn't: This time AM modulator.   

My friend, who likes to dig deep into old tube receivers, came to me with a question whether I knew how to generate a signal in between 100kHz up to some 20MHz or so, to test and tune those old beauties.

The obvious answer was to use for example a DDS for this, such as AD9834 (cheap 75MHz 10bit DDS). But then a question came, how to amplitude modulate the output?

Suppose I want to make such amplitude modulator, capable of working from 100kHz up to 30MHz (to cover LW up to the end of SW band). After spending some time with google and discarding all obviously foolish circuits, I have found this one, looking rather promising.


The differential approach through a transformer should have large rejection of the baseband signal at the output terminal. Well I, thought, let's have some fun building it. After messing a bit with the component values, it resulted in this:



The transformer is tri-filar wound 5mm toroidal core from N1 ferrite mix, 12 turns (a random number, that happened to fill the whole core circumference with those three 0.2mm wires).  ADT1-6T from Minicircuits would be a rather nice fit here, however the price is just nuts (and I do not have any, after all).

So after building the circuit, I tested its performance.  And become quickly disappointed.

First, the gain seemed lower than expected, well until I have found it is driven well in its saturation region, also producing nasty distorted output. I needed to keep input signal like 50dBm or below to have reasonably clean output.  Then it was able to produce gain of about 16-17dB, which seems good, but the P1dB point just absolutely horribly low.

I then thought, that this is just the property of the differential pair - it has to be driven with very low signal level, large distortion is produced otherwise. So okay, lesson learned there.

Second, the bandwidth. The circuit, as built, seemed it would not have much trouble working at up to some good tens MHz, even 100MHz - even with those plain simple BC847B I happen to have on hand. The gain decayed quite slowly, so these are probably good enough for those 30MHz. However at the low end of the band, say below 2MHz, a second and a third harmonic quickly appears at the output. I am not sure, but the transformer may be at fault? Probably core saturating? Uh, huh? Otherwise I can not explain this.

Third - the modulation. I tried modulating the output, however guessing above 30% of modulation depth, distortion is starting to come up quickly and I saw a lot of harmonics of the modulation signal.  I could not get any decent output above like 50% or so of the modulation depth. I haven't bothered to measure that accurately, it was obvious that is not enough.

So now the question is, how can the circuit be made better? Or should I use a completely different approach to it?
I would like to achieve at least 0 dBm P1dB, 0.1 - 30 MHz operation and say up to 90% modulation depth. What can be done about it?

A better transformer would probably get rid of distortion at the low end.
Using larger bias current (the circuit was originally biased at 2mA, I then upped the supply voltage from 5V to 10V and modified the bias to 12mA) helped a bit to make the P1dB higher, but by not much.
Would using emitter degeneration of the differential pair reduce distortion (increase P1dB)?
What other modifications or circuit solutions would get me to my goal?

[CJay]

https://www.analog.com/media/en/technical-documentation/application-notes/AN-423.pdf

[dzseki]

Or, you can simply use analog multiplier. Depending on your frequency needs you can find several models. AD835 for example goes up to 250MHz, with 2Vpp output.
Also you can read up on "Gilbert-cell" that is supposed to be the holy grail of the multiplier circuits regarding low distortion.

[David Hess]

A Gilbert-cell multiplier is one option and this used to be commonly done with the 1496 balanced modulator/demodulator which was sourced by practically everybody at one time or another.  Another way is to use a wideband operational transconductance amplifier.

[Yansi]

https://www.analog.com/media/en/technical-documentation/application-notes/AN-423.pdf

I know about this application note. But it is a rather nasty circuit of "we can do that too" type.  Parameters are strongly dependent on mosfet transconductance and gate threshold voltage. Due to manufacturing parameter spread of the mosfet, I consider this circuit unusable in practice. Linearity of such circuit is also very questionable.  I find absolutely unacceptable to have (at least) two values to trim for the modulator (offset and amplitude of the driving signal). Temperature drift would be a major one here probably too.

Or, you can simply use analog multiplier. Depending on your frequency needs you can find several models. AD835 for example goes up to 250MHz, with 2Vpp output.
Also you can read up on "Gilbert-cell" that is supposed to be the holy grail of the multiplier circuits regarding low distortion.

To use a $25+ four quadrant multiplier IC is not the suggestion I would expect, regarding the circuit in question above. Otherwise I could also generate the amplitude modulated signal digitally, using an FPGA and a DAC, or AD9957 digital IQ modulator or similar highly specialized IC. But this is not the goal here, but I appreciate your suggestion, especially of the gilbert-cell. I almost forgot about it.
 
So back to discrete semiconductors and my circuit. What could be done to tame it to perfection?

Would adding emitter degeneration increase the linearity?

What other modification could be done to increase its linearity, apart from completely changing the topology to such as the Gilbert cell?

Thank you David for that app-note, I am going to read through.

Thank you, Y.

[CJay]

https://www.analog.com/media/en/technical-documentation/application-notes/AN-423.pdf

I know about this application note. But it is a rather nasty circuit of "we can do that too" type.  Parameters are strongly dependent on mosfet transconductance and gate threshold voltage. Due to manufacturing parameter spread of the mosfet, I consider this circuit unusable in practice. Linearity of such circuit is also very questionable.  I find absolutely unacceptable to have (at least) two values to trim for the modulator (offset and amplitude of the driving signal). Temperature drift would be a major one here probably too.

It's crude but it works and actually sounds OK (I've tried it for a simple lashup, I also wanted to FM the DDS but that took more processor 'grunt' than I had intended to use)

YMMV of course but it seems to me it would be a lot simpler to design a suitable circuit to drive that pin at audio frequencies with the relevant de-emphasis rather than have to design something with flat gain and good linearity across 30MHz which seems to be your current problem.

[vk6zgo]

The best way to do this would to be to ditch your "wideband modulator", do your modulating at a fixed IF
frequency, then up or down convert the resultant AM signal as required.

The ubiquity of the "Superheterodyne" method bears witness to its advantages over direct modulation.
The first generations of analog TV transmitters had direct modulation at Channel frequencies ("wideband modulators" in the other sense.)
These were quite quickly replaced with "IF modulation" which gives you repeatable performance for any radiated frequency.

That said, old tube based signal generators managed reasonable modulation over a similar range to that which you require, with quite simple modulators.

[David Hess]

An MC1496 is less then $1.

[Yansi]

The best way to do this would to be to ditch your "wideband modulator", do your modulating at a fixed IF
frequency, then up or down convert the resultant AM signal as required.

The ubiquity of the "Superheterodyne" method bears witness to its advantages over direct modulation.
The first generations of analog TV transmitters had direct modulation at Channel frequencies ("wideband modulators" in the other sense.)
These were quite quickly replaced with "IF modulation" which gives you repeatable performance for any radiated frequency.

That said, old tube based signal generators managed reasonable modulation over a similar range to that which you require, with quite simple modulators.

And where is the advantage of that? I would still need to AM modulate a carrier signal, i.e. being able to multiply two signals. But now at least two times!

If one wants to make a synthesizer "the old way", I would need two rather high frequency oscillators, one fixed (your "IF") and one variable, to be able to synthesize 0.1 to 30MHz. Both oscillators would need be quite deep in the VHF region, to be able to produce a clean* frequency result in the range of 0.1 to 30MHz, after the final mixing.

*Clean means no significant harmonic content at the output. Doing this for a fix frequency TV channel, fine, but I think definitely not for a 300:1 range frequency generator.

An MC1496 is less then $1.

That rather comes as a surprise. O_o  Considering SA612 costs at least few times that and has almost the same guts inside.
Which comes to a nice idea: I do not have any MC1496, but plenty of SA612.  I could probably try a modulator using that one?

Also after a bit of digging, this came up: https://www.eevblog.com/forum/projects/ne602-ne612-double-balanced-mixer-amplitude-modulation-(am)-generation/ 

Questions about my original circuit still remain. I am not satisfied easily 

[Yansi]

https://www.analog.com/media/en/technical-documentation/application-notes/AN-423.pdf

I know about this application note. But it is a rather nasty circuit of "we can do that too" type.  Parameters are strongly dependent on mosfet transconductance and gate threshold voltage. Due to manufacturing parameter spread of the mosfet, I consider this circuit unusable in practice. Linearity of such circuit is also very questionable.  I find absolutely unacceptable to have (at least) two values to trim for the modulator (offset and amplitude of the driving signal). Temperature drift would be a major one here probably too.

It's crude but it works and actually sounds OK (I've tried it for a simple lashup, I also wanted to FM the DDS but that took more processor 'grunt' than I had intended to use)

YMMV of course but it seems to me it would be a lot simpler to design a suitable circuit to drive that pin at audio frequencies with the relevant de-emphasis rather than have to design something with flat gain and good linearity across 30MHz which seems to be your current problem.

Yes, crude and unusable. Almost zero repeatability. I do not doubt it working, but the parameter spread is what prevents one to use the circuit reliably. Also, decreasing the output amplitude of the DDS will I think worsen the SNR (and probably other properties too). It is I think always good to keep DAC output levels maximum possible, (not only) to reduce cross-talk from the digital circuitry. Not that it would matter so much with such cheap 10bit DDS, but as a matter of principle.

Gain (flatness) is non-issue here. Linearity is on the other hand a big one. 

[vk6zgo]

The best way to do this would to be to ditch your "wideband modulator", do your modulating at a fixed IF
frequency, then up or down convert the resultant AM signal as required.

The ubiquity of the "Superheterodyne" method bears witness to its advantages over direct modulation.
The first generations of analog TV transmitters had direct modulation at Channel frequencies ("wideband modulators" in the other sense.)
These were quite quickly replaced with "IF modulation" which gives you repeatable performance for any radiated frequency.

That said, old tube based signal generators managed reasonable modulation over a similar range to that which you require, with quite simple modulators.

And where is the advantage of that? I would still need to AM modulate a carrier signal, i.e. being able to multiply two signals. But now at least two times!

If one wants to make a synthesizer "the old way", I would need two rather high frequency oscillators, one fixed (your "IF") and one variable, to be able to synthesize 0.1 to 30MHz. Both oscillators would need be quite deep in the VHF region, to be able to produce a clean* frequency result in the range of 0.1 to 30MHz, after the final mixing.

*Clean means no significant harmonic content at the output. Doing this for a fix frequency TV channel, fine, but I think definitely not for a 300:1 range frequency generator.

An MC1496 is less then $1.

That rather comes as a surprise. O_o  Considering SA612 costs at least few times that and has almost the same guts inside.
Which comes to a nice idea: I do not have any MC1496, but plenty of SA612.  I could probably try a modulator using that one?

Also after a bit of digging, this came up: https://www.eevblog.com/forum/projects/ne602-ne612-double-balanced-mixer-amplitude-modulation-(am)-generation/ 

Questions about my original circuit still remain. I am not satisfied easily 

The "old way" will work perfectly well if you have several ranges.
No, you don't have to Amplitude Modulate a carrier more than once.

Up & down conversion  of an already modulated carrier is a lot easier.
It is done all the time in Amateur Radio HF Transmitters, which commonly cover a  range of frequencies from 1.8 Mhz to 30 MHz, albeit, in a number of selectable bands ( some can be modified for continuous coverage, although that is illegal)
Not quite your 0.1 to30 MHz but close.

Why do you have to have it continuously tuneable?

Unfortunately, in the real world, the only way you can get rid of harmonics & spurii, is to use selective tuned circuits, or traps, &/or  take smaller "bites" of bandwidth where you can use Band Pass Filters.

[IconicPCB]

As  first step towards a better balance, try a 1K pot in the emitter circuit of Q1 and Q2 connect the wiper to the collector of Q3.
Apply the signals and try adjusting the balance between the two transistors.

Also swap C6 and C5

[Yansi]

The "old way" will work perfectly well if you have several ranges.
No, you don't have to Amplitude Modulate a carrier more than once.

Up & down conversion  of an already modulated carrier is a lot easier.
It is done all the time in Amateur Radio HF Transmitters, which commonly cover a  range of frequencies from 1.8 Mhz to 30 MHz, albeit, in a number of selectable bands ( some can be modified for continuous coverage, although that is illegal)
Not quite your 0.1 to30 MHz but close.

Why do you have to have it continuously tuneable?

Unfortunately, in the real world, the only way you can get rid of harmonics & spurii, is to use selective tuned circuits, or traps, &/or  take smaller "bites" of bandwidth where you can use Band Pass Filters.

Old way will probably work, with MINIMUM number of 2 multipliers (one for modulator), other for mixing the carrier with another LO. I will also need two low pass filters with steep roll-offs for the two, to produce harmonic-free signal from 0.1 to 30MHz.  Generating signal this way is MUCH more complex circuit.

You should perhaps state some facts why it is better, because I do not see anything better in that and already provided facts why it is not.

To restate these facts:

Old school frequency synthesis uses two rather high frequency oscillators, mixed together to produce a signal with low harmonic content. All harmonics of the oscillators must be filtered out, then mixed together with a highly linear mixer (so no new harmonics are created) and then again low pass filtered.

This way one needs a minimum of TWO oscillators: One fixed frequency, one continuously variable one, over the required output range. The fixed LO needs to have the AM modulator (to satisfy your requirement of modulating a fixed IF frequency), in this case the modulator will need to work deep in the VHF region, although fixed frequency.
Then I need minimum two analog multipliers, one for the AM modulator, one for the final mixing.
I also need two VHF low pass filter and one output low pass filter (brobably not that much needed, but better to have one).
Not counting the numerous buffer amplifiers and places requiring tweaking of signal levels and obligatory impedance matching, also the need for using a high frequency DDS or fractional PLL for the variable LO, to be able to obtain stable frequency output and continuous tuning step of say 10Hz across the whole 30MHz output range.

What do you see better in this approach, compared to the one at the bottom of the image?

What is the argument to not have a wide-band modulator? Apart that it could have a tuned LC tank at the output, to suppress a little bit of upper harmonics (still it would have to be low pass filtered). Still you need an analog high frequency multiplier, with good linearity.
You may probably argue with gain variation over frequency, but you will have that one even in your complex synthesizer topology, due to the variable LO.  Output signal level is however fixed very easily using an ALC loop.

I do not see any advantages of the complicated way you are presenting.

Why do have I to have it continuously tuneable?  Because I am trying to design a simple HF generator from 100k to 30MHz. Not an LO source for an amateur transceiver with just few very narrow bands. Originally, as stated in my first post in this thread, this will serve as a signal source for testing and fixing LW/MW/SW tube receivers. The frequency ranges available on these are pretty wide compared to an amateur transceiver.

As  first step towards a better balance, try a 1K pot in the emitter circuit of Q1 and Q2 connect the wiper to the collector of Q3.
Apply the signals and try adjusting the balance between the two transistors.

Also swap C6 and C5

So in fact you are wanting me to add emitter degeneration (that's what I am asking from the beginning). Will try it.

I will however not swap the two capacitors (C5, C6), only increase capacitance of C6. I think I know where you are going with that. However that still won't fix poor P1dB/IP3 of the modulator (emitter degeneration might?), will just reduce cross-modulation of the bias point from the modulating signal input.

[BigMark]

You might be better off researching 'Heising Modulation' (High Level Modulation) which was popular in the 1930's with various tube transmitters. Tended to use a choke or low pass filter to isolate the AF and RF stages. I think your just trying to drive the RF stages of an old AM receiver, so you don't need to run much power and therefore is little need to over complicate the stages with exotic filtering.

A quick look I found this website which might be a guide.
https://www.w8ji.com/Heising%20modulation.htm

For the choke you could just use a low pass inductor/capacitor filter. Inductor in series for DC pass. A circuit using FET's can be found in this book at the top - page 652. Also see attached image.
https://books.google.co.uk/books?id=uoj3IWFxbVYC&lpg=PA652&ots=11rUNkQAsS&dq=heising%20modulation&pg=PA652#v=onepage&q=heising%20modulation&f=false

Using a mixer like someone proposed is low level modulation which is popuilar with transceivers and TV transmitters. Probably over complicated for your needs. TV transmiters had major lineartiy requirements which required video level and IF pre-correction and transceivers tended to use many common stages to reduce the size/scale of the equipment. So AM and FM shared many stages.

[Yansi]

This exactly would not work as I intend it to. This will produce significant harmonic content and low pass filter will not correct that - there is no way a single low pass filter would get rid of higher harmonics, considering the amount of tuning range (300:1) I need.  It would need multiple (not two, or three, but likely tens of them) to be able to get to the level of signal clarity a DDS produces.

I think the wide-band Gilbert cell approach is so far the easiest and best one. I just need to etch a board for the NE612 to have a go with it, until the "proper" MC1496 will arrive.

I just need to make that analog multiplier to perfection, that's all what's needed.  I have added the emitter degeneration resistors and it did quite the expected: Significantly decreased gain (up to the point it has only attenuation, like about 15dB or so with 2x 470R) and increased its capability to deal with more dynamic signals - which was also expected.

However the modulation characteristic of the single diff-pair circuit is still very poor. I can't get over few % modulation depth without introducing significant distortion in the modulated output.  I do not know what's wrong, probably circuit biasing? Or it is just the limit of this simple circuit, that could be fixed with the Gilbert cell?

[Wolfgang]

Hi Yansi,

another chip you may try is the good old NE602/612, also a Gilbert cell multiplier up to more than 100MHz.

[Yansi]

Hi Wolfgang, have you not probably read the thread, I am already etching a test board for it. But thanks for the tip 

[Wolfgang]

Sorry, your last post slipped my eyes. Much success ! 

[David Hess]

An MC1496 is less then $1.

That rather comes as a surprise. O_o  Considering SA612 costs at least few times that and has almost the same guts inside.

The 1496 balanced modulator/demodulator is more of a basic building block with many configurations and applications so demand is higher yielding lower costs.  Practically anything else is more specialized.

Another option in the past would be to configure a couple of transistor arrays as essentially a 1496.  You could also build a 1496 or half of a 1496 out of matched discrete transistors.

[Yansi]

I have never heard before and never seen a MC1496. Probably not so popular IC around here. But SA6x2 is a pretty common one.

So, I have built a simply stupid circuit using the SA602 to make AM modulator out of it.  I am feeding a little offset current to one of the inputs, to have a bit of carrier leak through (otherwise I would get a pretty good DSB modulation).

It works very good, however only with very low input signals. The output is also extremely weak. Can't get more than -42dBm carrier level output, otherwise significant distortion occurs at the output.

The input carrier level (into the OSC-B pin) is equivalent to -15dBm,  which is around 40mVrms
the modulation input level is 30mVp (21mVrms).
And I am getting output carrier level of -42dBm, the overall output level is 3.2mVrms.  Jeeez, that's fu*ng low!

If I try to push the levels, unacceptable distortion occurs.  To be honest, I do not have much like for such low signal levels

First I think the ultra-weak output is caused by my non-ideal output circuit configuration. I have connected a 1:1 RF transformer directly at the SA602 output pins.

If I now think about it (apart from I should stop copying stupid designs from the web), the differential output resistance of the SA602 is 3kohm, as each collector is internally loaded with 1.5k resistor to V+, and the circuit is working in constant current mode - so the output resistance is defined mainly by the internal load resistor of 1.5k for each output, making 3k output resistance differentially.  (Yes, I mean resistance, ignoring the parasitic capacitances, which at my low frequency of few MHz are probably insignificant to the problem)

So this output resistance (or impedance, whatever you call it) makes a nasty voltage divider, that eats my signal level - as of course, as a load sits my spectrum analyzer with 50 ohm input.

If I think about it, using a center tapped transformer with the center tap connected to V+, as the one in my previous circuit, the internal 1.5k resistors become parallel to the transformer, not in series with it, which should increase the signal level substantially, if I am correct.

//EDIT: Using center tapped transformer made the level only few dB higher, like a 3dB or so (now overall output level measured is 5.6mVrms). Still too low.

[Yansi]

Well silly me, I could just use the transformer to match the output impedance.  But the ratio is rather nasty 7.75:1.

Still, using such transformer (31 to 4 turns) did not improve it by much, output level still only circa 11mVrms (carrier at -33dBm or around that)

So very very small levels everywhere, I do not like that really.

[Wolfgang]

Yansi, did you have alook at the Philips/NXP app notes about the NE612 ? I think they have several input/output matching circuits there, IMHO.

[nugglix]

"If the required LO is beyond oscillation limits, or the system calls for an external LO, the external signal can be injected at Pin 6 through a DC blocking capacitor.
External LO should be at least 200mVP-P."

From page 4 of the attached data sheet.
40mVrms seems to be a bit below the required voltage.

[Yansi]

I call NO on that required datasheet voltage, as at such level, I get significant second and third harmonic of LO appear at the output. 3rd harmonic is only like 30dB bellow the fundamental.  That is unacceptable.  Conversion gain may be higher, though, but linearity suffers significantly.

At the level of -20 to 15dBm of LO level, the second and third harmonics at the output are negligible, at least for my spectrum analyzer and the level of fundamental (+ the 2 side-bands of modulation) does not have that much lower level.   

[Wolfgang]

What you could do to keep the NE612 in the "linear" range and still have a useful output amplitude is to add a differential video amp at the NE612s output.
How many dBms at the output do you really need ?