Frequency multipliers

[Enthalpy]

Greetings to all HF enthusiasts!

 

This shall be a frequency multiplier, but beware I haven't tested it (and won't in a foreseeable future). It will multiply, but whether it improves over existing diagrams remains to be seen.

 

 

This diagram is to harvest an odd harmonic; connect the collectors together at one end of the coil to harvest an even harmonic.

 

A balanced mixer could be built similar to the multiplier, with the LO input between the input secondary's middle point and the ground; adding two PNP would make a doubly-balanced mixer. I doubt such a mixer has advantages over a diode bridge.

 

-----

 

Unusually, the multiplier has bipolars in common base without a power supply. All power comes from the input signal.

 

• The bipolars conduct at some 0.7V input voltage but drop only about 0.4V; this behaviour is more brutal than a diode, and shall transfer a bigger proportion of the input power more concentrated in the harmonics, so I hope losses are reduced as compared with a diode frequency multiplier;
• The emitter current determines essentially the collector current, so imbalances between the transistors are less important than with the traditional two-transitors multiplier; operation nearer to Ft improve as well;
• The output power is well defined and should depend less on the input power than with diodes or traditional transistor diagrams.

Feel free to match the transistors by Vbe, Ft, capacitance... or use a matched pair in a case. I estimate that the fully balanced connection of the transformers, with a cold point at the middle, brings more to an odd-only or even-only output spectrum.

 

Stability isn't guaranteed, as the input signal brings power to the circuit. Beware many recent bipolars accept only a common-emitter configuration. I'd take only the necessary Ft, and wouldn't put the output tuning capacitor near to the collectors.

 

Marc Schaefer, aka Enthalpy

 

=================================================

 

This one relies on the ability of either gate of a tetrode MOS to pinch the drain current, which here passes by zero twice in an input period. Beware again, I haven't tested this circuit. It is meant for even multiplication only.

 

 

Because the two gates aren't identical, I've put two transistors in a symmetrical circuit to minimize the odd harmonics.

 

Injecting the signal in the upper gate isn't perfect for gain nor stability. A cascode would improve that, possibly shared by both tetrodes, using a bipolar if stable in common base, a single gate HF MOS if you find one, or a dual gate MOS. Bias as normally the second gate would be. Especially then, the multiplier would accept a small input power.

 

Feel free to add in each source a capacitor and a resistor, or much better, a precisely matched integrated current source. This is interesting with other diagrams as well.

 

Replace the MOS by MESFET, HEMT and other variants ad libitum.

 

Marc Schaefer, aka Enthalpy

 

=================================================

 

The second frequency multiplier would ideally use a matched double pentode component, still to be made, like this:

 

 

The third gate would make the cascode without cabling penalty.

 

Two transistor on one chip would be matched by fabrication. Or even, have four transistors cabled internally in pairs in a cross pattern, as in op amps, if this improves further.

 

MESFET, HEMT and the like can improve over MOS.

 

Maybe these component and circuit will be used first as part of a more complex integrated circuit and be made available later just as a matched pair.

 

I have found no use whatsoever for a filament, alas.

 

Marc Schaefer, aka Enthalpy

 

=================================================

 

The same matched double pentode would make a nice mixer.

 

 

Tetrode MOS are already excellent mixers, whose intermodulation is second to diode rings only, provide gain instead of loss hence noise, and need little LO power. A symmetric circuit can only improve intermodulation.

 

Whether the local oscillator best drives the gates 1 or 2 against intermodulation is unclear; against injection, gates 2 are better. Bias should also be optimised against intermodulation. Again, the sources may have a capacitor and a resistor or a current source.

 

In fact, an even more symmetric circuit would be even better. Put four transistors on a single chip, and at one pair, swap the inputs of either the signal or the local oscillator; bring the new drains to the opposite end of the now symmetric output transformer. Best described with words, as the diagram is a mess. Refer to the SO42's input swaps if you wish.

 

Maybe the four transistors in cross pattern suffice to compensate all linear geometric gradients in the production process; cable 16 transistors on the chip if not. In any case, I wish that such a mixer component remain usable as a frequency multiplier as well.

 

Here also, MESFET, HEMT and the like can replace MOS. Low-capacitance outputs should be preserved, as the output signal is fast at some uses.

 

Marc Schaefer, aka Enthalpy

Edited December 3, 2012 by Enthalpy

[Enthalpy]

The doubly balanced mixer has gotten a half-way legible diagram; click to enlarge.



The cabling mess better happens within the chip that carries the matched transistors, like in the old SO42. Replace the MOS by other multi-gate FET as you like. The third gate saves some power from the local oscillator if the intermediate frequency doesn't differ much; its bias acts on intermodulation.

Marc Schaefer, aka Enthalpy

[Enthalpy]

With valves as with Jfet, symmetric preamplifiers were built to improve linearity at receiver input. I don't understand the benefit, since this reduces only the even nonlinearity which isn't a big drawback at a preamplifier, where transmodulation results from odd distortion. In case I got it wrongly, the matched pair of multigate transistors would be usable as a preamplifier as well.

 

A high impedance shared by the sources of transistors working symmetrically reduces further the even nonlinearity. Multigate transistors are more stable here as well. The impedance can be a coil, a resistor, a current source if not capacitive... this latter being more likely if integrated in the same chip as the HF transistors.

 

This can be interesting for an odd frequency multiplier, and also for the doubly balanced mixer with four transistors sketched above (not for the two-transistors mixer), where all four sources can be tied together (possibly through capacitors) to share a common constant current, say from current source(s). It would linearize further the circuit, improving the intermodulation behaviour at the signal input.

 

Marc Schaefer, aka Enthalpy

[Enthalpy]

This frequency tripler starts like a distributed amplifier, with a circuit to provide signals 120° apart, and drives three nonlinear components whose outputs cumulate without phasing in a filter. This isn't uncommon, and attenuates the harmonics 1, 2, 4... when the third is wanted.

Usual circuits would drive components, for instance bipolar transistors, to a strongly nonlinear amplitude. The small conduction angle lets the output current contain usable harmonics, in addition to the strong fundamental.

Here I propose to use Fet with 3 or 4 gates connected to the three phases. Any gate can cut the drain current, which passe hence by zero three times per input period. This squeezes the fundamental amplitude in the drain current, and it also reduces the power consumption.

The fourth gate is optional. It enables a smaller input power, and with the proper bias, it reduces the even harmonics.

Three paired transistors, whose gates swap the input phases, symmetrize the output current, squeezing further the unwanted harmonics.

I've drawn a three-phase ring resonator at the input, downstream the phaser. This one uses identical components and equalizes the amplitudes and phases that are sensitive to tolerances at the phaser. Additional ring resonators are possible.

Two resonating circuits can suffice at the output, three are easier at a tripler because the filter must squeeze the fourth harmonic near the third. Traps would be sensible.

The circuit could multiply the frequency by 5 with more transistors and gates, but the filters become uneasy and the output signal weak, as usual.

For Hf, Vhf and Uhf, a Mos silicon chip could carry the three multigate transistors already connected, for use with an external filter. For microwaves, a chip of adequate semiconductor could integrate faster Fet and the passive components, as lumped constants, or at higher frequencies as microstrips and rings.

Untested, but this one can only work.
Marc Schaefer, aka Enthalpy

[Enthalpy]

Chains of frequency doublers, integrated on a chip, could be catalogue parts.

To achieve frequencies above ~500MHz from a quartz or Saw resonator, Pll are more fashionable, but:

• Last time I searched for a prescaler at 40GHz, I didn't find any.
• Pll introduce phase noise, waste power and synchronize slowly on a new frequency.

Frequency doublers pick the second harmonic from the nonlinear element with a frequency filter, usually narrowband and centred on each design's need. Careful design could make a larger passband and keep the rejection, especially with transmission zeros. The filter must pass the second harmonic and block the third and the fundamental, so the input frequency can't vary much.

• 1 to square root(2) ~1.414 input frequency band is (too) difficult: pass the second harmonic from 2.000 to 2.828 but block the third above 3.00, demanding 1.061 selectivity despite components tolerances and finite filtre steepness.
• 1 to cubic root(2) ~1.260 is reasonable: pass 2.000 to 2.520, block at 3.000 and beyond, selectivity 1.191.
• 1 to fourth root(2) ~1.189 is easier: pass to 2.378, block at 3.000, selectivity 1.261.

A doubler chain chip would usefully expose the inputs and outputs of each stage, so the user chooses where his signal enters and leaves the chain according to the frequencies. The user then connects outside the chip the stages together, or sets control pins to do it internally and save buffer power. He may also shut down the unused stages. If the input frequency band is 1 to cubicroot(2), three catalogue chips cover all needs.

    [DoublerChain.png]

A fast semiconductor, GaAs or more recent, can integrate the active and passive components up to 40GHz or 95GHz+ for instance. The passive components for the stages below few GHz reside better outside the chip, which can still provide the transistors. Alternately, a distinct chip can carry only wideband active components for a bunch of doublers and triplers, on a cheaper process like Cmos/Si. A hybrid circuit can carry passive components outside the chip and be still a catalogue part. On-chip or on-hydrid tuning would be useful if possible, at biasses to minimize the unwanted harmonics, and even better at the filters - not only for the present chain.

The doubler circuit I proposed here on 03 December 2012 saves power and eases the harmonic filter. Expect more components at the selective wideband filter than the two resonating circuits I drew for the narrow passband. The tripler I proposed here on 18 May 2019 would be conceivable but it restricts the bandwidth even more, needing more catalogue chips to cover all frequencies, so I'd put some rather at an aperiodic (part of a) chip.

One catalogue chip could also carry all the doubler chains needed to cover all input frequencies and let the user drop away and switch off the unecessary ones, through cabling or by setting control pins.

Marc Schaefer, aka Enthalpy

[John Cuthber]

I wonder if anyone has ever set up a chain of frequency multipliers (doublers or whatever) to convert the very accurate 60 KHz signal used by radio controlled clocks up to something in the MHz region for calibration/ checking of frequency meters.
You can get multipliers designed for this but they are PLL based.

A sequence of accurate 60, 120, 180... KHz test frequencies might be useful.

A trippler gives you 180 KHz and a quintupler gives you 300 KHz

 

Any thoughts?

[Carrock]

1 hour ago, John Cuthber said:

I wonder if anyone has ever set up a chain of frequency multipliers (doublers or whatever) to convert the very accurate 60 KHz signal used by radio controlled clocks up to something in the MHz region for calibration/ checking of frequency meters.
You can get multipliers designed for this but they are PLL based.

A sequence of accurate 60, 120, 180... KHz test frequencies might be useful.

A trippler gives you 180 KHz and a quintupler gives you 300 KHz

 

Any thoughts?

Most (digital) frequency meters can also be configured to measure time. A divide by 60,000 or more device would at least be simpler than a multiplier. Noise and jitter in the MSF signal would be a problem; a very large division ratio would reduce those but maybe not enough.

Getting high accuracy is the big issue whatever you do.

[Enthalpy]

Did I see such a circuit in Elektor or Radio Plans, to multiply the frequency of a reference transmitter? It was just after the last ice age, as the DCF77 antenna emerged from the melting shelf, so I don't remember quite accurately.

Frequency *5 is perfectly reasonable with a transistor and a filter if you can adjust the resonant circuits, I did it a couple times. For series production, it demands serious precautions. Or individual tuning too.

Nowadays, Pll and Dds are simpler, except for special needs. The standard way to get an accurate second is Gps. Horribly more complicated, but available ex stock.

It resembles the electronic bubble level someone asked for elsewhere. He wanted to build the sensor, I suggested to put a webcam in front of a usual bubble level, plus image analysis software on a Pc. If you don't need to develop a webcam and a computer only for this, it can be considered. And, gasp!, the guy was happy to have a "simpler" solution than developing the sensor. Software has pretty much killed the fun in electrical engineering.

[swansont]

15 minutes ago, Enthalpy said:

 The standard way to get an accurate second is Gps.  

There are exceptions to this, of course.

 

[Sensei]

1 hour ago, Enthalpy said:

The standard way to get an accurate second is Gps.

@swansont

..I made Android application for my smartphone, which was connecting to GPS satellites, and getting their data. From it calculated latitude and longitude and altitude and saved to CSV (Comma Separated Values) file format. It can be imported by OpenOffice SpreadSheet or Excel, and special created for this project Google Maps PHP script. And started walking in the city with app turned on. While I was on the surface precision was +- few meters on the Google map script. But as soon as I went down the stairs near river, it all started malfunctioning. I was recorded being once on the right side of the river, the second record of data (took just a few seconds later), on the second side of the river (and between 200 to 400 meters away from where I used to be in the reality). It jumped this way entire time during walking close to the river. Why? Because there are massive flood walls (concrete floodbanks) 5+ meters tall. Signal from satellite cannot reach device in the straight line, but bounces from flood walls and reflection of signal arrives with higher delay than it should be.. and software has to analyze database to get ride of such incorrect entries..

What is accurate for you might be not accurate enough for somebody else.

Edited May 20, 2019 by Sensei

[Strange]

1 minute ago, Sensei said:

What is accurate for you might be not accurate enough for somebody else.

200 metres is less than a microsecond. That is probably accurate enough for most people. 

And if you need higher accuracy, use a GPS receiver that has a clear view of multiple satellites.

[swansont]

2 hours ago, Sensei said:

@swansont

..I made Android application for my smartphone, which was connecting to GPS satellites, and getting their data. From it calculated latitude and longitude and altitude and saved to CSV (Comma Separated Values) file format. It can be imported by OpenOffice SpreadSheet or Excel, and special created for this project Google Maps PHP script. And started walking in the city with app turned on. While I was on the surface precision was +- few meters on the Google map script. But as soon as I went down the stairs near river, it all started malfunctioning. I was recorded being once on the right side of the river, the second record of data (took just a few seconds later), on the second side of the river (and between 200 to 400 meters away from where I used to be in the reality). It jumped this way entire time during walking close to the river. Why? Because there are massive flood walls (concrete floodbanks) 5+ meters tall. Signal from satellite cannot reach device in the straight line, but bounces from flood walls and reflection of signal arrives with higher delay than it should be.. and software has to analyze database to get ride of such incorrect entries..

What is accurate for you might be not accurate enough for somebody else.

Sorry, I was being a bit snarky.

USNO, which where I work, is the source of time for GPS. IOW, the standard way for us to get time is to measure it locally, since that measurement is better than what you can get from GPS.