Frequency Converters: Specifying the Whole Module

A mixer translates frequency. A converter does everything that has to be true for the translation to be usable: it rejects the band that would otherwise arrive on top of the wanted one, it supplies an LO clean enough not to spoil the signal, and it hands the result to the next stage at a level and in a passband something downstream can accept. That is why converters sit in this components section as their own family rather than as an accessory to the mixer page.

The practical consequence is that most of a converter datasheet describes parts that are not the mixer.

What is actually in the box

The satellite world names the assemblies plainly, which makes it a good place to see the pattern. A block upconverter is described as an outdoor unit that upconverts an L-band IF, typically 950 to 1450 MHz, to the transmit frequency, and its published contents are a mixer, an LO synthesizer locked to a 10 MHz reference, an image-reject filter and a power amplifier. The receive-side counterpart, the low-noise block, is the same idea with an LNA in place of the PA.

Four blocks, four different specifications, one part number. A general-purpose lab converter, an EW receiver front end and a point-to-point radio module differ in which of the four dominates, not in the list.

The frequency relationship itself is arithmetic: for the upconverter above, the output is the LO plus the IF. Which sideband you keep, and which one you have to remove, is the subject of upconverter vs downconverter and the frequency-plan identities in RF, IF and LO explained.

The mixer sets a floor, not the result

A passive diode mixer loses power by construction. The theoretically optimum conversion loss for a passive diode mixer is 3.9 dB, with typical values between about 4.5 and 9 dB depending on topology and bandwidth; 6 to 8 dB is common for standard double-balanced parts, and active mixers can convert with gain instead.

None of that appears directly on a module datasheet, because the module quotes conversion gain — the mixer’s loss and the amplifiers’ gain, netted. Two things follow. A converter with 20 dB of conversion gain still contains a mixer working at its own loss, so the mixer’s compression behaviour is still in there. And the module’s noise figure is set mostly by whatever comes first: put the loss ahead of the amplifier and it lands on the noise figure directly.

Port isolation is the other mixer-level number that reaches the outside world. Published ranges put LO-to-IF isolation at typically 20 to 30 dB and RF-to-IF at 25 to 35 dB, which is why the IF filter in the diagram is not optional decoration: LO leakage at the IF port can saturate the IF amplifier, and it degrades conversion-loss flatness before it does anything as obvious as producing a visible tone.

Large-signal behaviour is tied to the LO as well. As a rule of thumb, the 1 dB compression point sits 4 to 7 dB below the minimum recommended LO drive level, so a module built around a low-drive mixer has a lower ceiling regardless of how much IF gain follows it.

Spurious products are a plan, not a fault

A mixer produces output at combinations of the LO and RF harmonics, and only the first-order product is wanted; the rest exist because real diodes do not switch ideally. Balanced topologies suppress them structurally rather than by filtering: relative to a single-diode mixer, a single-balanced design removes about half the intermodulation products, a double-balanced design as much as 75%, and triple-balanced designs further still.

What matters for specification is that every remaining product has an address. Its frequency is a fixed combination of your LO and RF frequencies, so it moves when the frequency plan moves. A spur that lands inside the IF passband at one LO setting may be harmless at another, and no amount of module quality removes it — which is why converter datasheets quote spurious levels against specific offsets rather than as a single number.

Third-order intermodulation is the case that cannot be filtered at all, because the products land too close to the wanted IF to separate, and they rise three times faster than the signal that causes them. That is a dynamic-range budget, decided before the module is chosen.

Where the image goes

Every downconversion has a second input frequency that converts to the same IF. There are three ways to deal with it, and a converter datasheet is partly a statement of which one the vendor picked.

Filter it. A preselector ahead of the mixer removes the image band before it reaches the mixer. This works well when the IF is high, because the image is then far from the wanted signal. It gets expensive as the IF falls and the two move closer together. When the receiver has to tune, the filter has to tune with it, which is the case for YIG-tuned filters; when the band is fixed, a cavity is usually the better filter.

Cancel it. An image-reject mixer splits the signal into quadrature paths and combines them so the image lands at a port that can be terminated. The rejection then depends on how well matched those paths are: 10° of phase unbalance and 1 dB of amplitude unbalance, described as easily achieved, give about 20 dB of image rejection, and a published wideband part measures typically 20 dB across most of its band, better than 25 dB from 40 to 52 GHz. Useful, integrated, and not on its own a substitute for preselection when the interferer is strong.

Plan around it. Choosing an IF that puts the image where nothing transmits costs nothing in hardware and is the first thing to try. It is also the option that disappears once someone else has chosen the module’s IF for you.

The LO is the other half of the converter

The LO is the block that most often decides whether a converter is acceptable, and the one least visible in a block diagram. Its phase noise is impressed on the converted signal, and no filtering downstream removes it: for the mechanism and how to read L(f), see phase noise. Where the LO comes from is a component decision in its own right — a synthesizer when frequency accuracy and step size matter, a YIG-tuned oscillator when the range is an octave or more and close-in noise has to stay low.

Three LO properties reach the module specification directly: spectral purity, which becomes EVM or reciprocal mixing; step size and tuning range, which decide the frequency plans available to you; and settling time, which decides how fast the converter can be retuned. A radar or EW receiver that hops is buying the third one as much as the first.

Reading the datasheet

Parameter	What it describes	What it constrains	Where it comes from
Conversion gain	Net gain of the assembly	Level presented to the next stage	Mixer loss netted against amplifier gain
Noise figure	Assembly noise	Sensitivity	Dominated by whatever is first in the chain
Image rejection	Suppression of the second input band	Which off-channel signals reach the IF	Preselector, image-reject mixer, or IF choice
Isolation and LO leakage	LO power escaping to other ports	IF stage saturation, radiated interference	Mixer balance plus filtering
Spurious, in dBc at stated offsets	Unwanted mixing products	Which frequency plans are usable	LO and RF harmonic combinations
P1dB and IP3	Large-signal behaviour	Upper end of dynamic range	Mixer topology and available LO drive
LO phase noise	Spectral purity of the LO	EVM, reciprocal mixing, close-in noise	The LO source and how it is locked
Gain flatness	Gain against frequency	Equalisation and multi-carrier balance	Filters and amplifier response

Ranges quoted in this article describe the published device class, not a model specification. Read every number against the LO drive, temperature and port conditions stated on the datasheet.

Which parameter binds, by application

Application	Binding constraint	What that forces
Wideband receiver or EW front end	Image and out-of-band interference	Tracking preselection and a wide, quiet LO
Spectrum analyser front end	Spur-free dynamic range	High first IF and a frequency plan chosen around spur addresses
Radar receiver	Retune or hop time	LO settling, not conversion gain
Satellite uplink chain	Output power and frequency stability	Locked reference and specified gain flatness
Point-to-point radio	EVM of a modulated carrier	LO phase noise budget and IF flatness
Laboratory band translation	Level accuracy and repeatability	Known conversion gain over temperature, and headroom

One constraint per row is deliberate. When two of these bind at once the answer is usually a different frequency plan, not a better module.

The radar row is worked through in radar RF component selection, where the waveform, not the part list, decides which constraint binds first.

Integration and test

Filtering interacts with the mixer rather than sitting politely after it. A reflective filter returns out-of-band energy to the mixer port, which shows up as conversion-loss ripple and as spurs that were not in the plan; the standard mitigations are to place the filter close to the output, and to use a terminated diplexer or a small pad where the interaction persists. Inside a module this has already been done for you, well or badly, and it is one of the things you are paying for.

Three checks are worth doing on receipt, because they catch the mismatches a datasheet cannot:

• Measure conversion gain at the LO drive and input level you will actually use, not at the datasheet’s condition.
• Sweep the LO across your real plan and look for spurs in the IF passband, rather than trusting a single quoted offset.
• Check the IF port for LO leakage with the IF amplifier in circuit, since that is where isolation turns into compression.

When the module you want is discontinued — common for converters, which are often built in small volumes for one programme — the sourcing route and its verification steps are covered in obsolete RF component sourcing.

Compare modules under one frequency plan

A converter comparison only becomes useful when every candidate is placed in the same plan. Start by writing the wanted RF band, the intended IF, the LO range, the maximum wanted level and the strongest expected blocker. From those five entries, calculate the image, the unused mixing product and the frequencies at which the IF filter must reject energy. This prevents a familiar mistake: selecting a lower-loss module whose fixed IF puts the image directly in a band the front end cannot tolerate.

Then record the conditions behind each headline number. Conversion loss is normally quoted for one LO drive and one RF level; change either and the number can move. Third-order intercept and compression are only comparable when the tones, output frequency and measurement reference plane are the same. Isolation deserves the same discipline, because an acceptable LO-to-RF leak at a bench connector can become a radiated or compression problem once an amplifier and antenna are connected.

Finally, treat the converter, LO and filters as one acceptance item. Measure gain, spurious response and leakage with the real source, the real IF bandwidth and the filters in their intended order. If the plan is still flexible, moving the IF is often cheaper than buying a more elaborate mixer module. The RF, IF and LO guide is a useful final check before a part number becomes a drawing.