# Why is superheterodyning better than direct conversion?

What is the advantage of having the extra stage of IF, extra local oscillator and extra filtering stage and amplification. It just seems like more work and circuitry. Surely the direct conversion to baseband and the intermediate conversion to baseband are both going to go through the same passband filter at the end of it that isolates the original frequency range?

I see answers such as 'it allows common circuitry to be used rather than separate circuitry for each frequency', basically:

Without using an IF, all the complicated filters and detectors in a radio or television would have to be tuned in unison each time the frequency was changed, as was necessary in the early tuned radio frequency receivers

But I don't understand because they could all be tuned to the baseband signal for instance rather than the IF, eliminating the problem described.

I also see:

So a narrower bandwidth and more selectivity can be achieved by converting the signal to a lower IF and performing the filtering at that frequency

But why not perform the filtering only at the baseband frequency IF=0. That is a lower IF, right. Does the addition of the IF passband stage improve the sharpness of final passband as opposed to doing it directly?

The only disadvantage I can understand is the leaking local oscillator and doesn't this happen more at higher frequencies? An IF could allow the first LO to oscillate much slower.

• I don't know enough to say for sure, but consider the size difference between low-frequency and high-frequency inductors and capacitors of the same impedance. The high-frequency ones are a lot smaller. Commented Apr 13, 2019 at 14:38
• Opposite sideband rejection is one issue (unless you do the quadrature thing a direct conversion set does not have any!), while an intermediate LO makes this relatively easy. The other elephant in the room is the massive dynamic range present at the mixer output, it is hard to build a low frequency LPF with steep skirts and a 100dB++ dynamic range. Commented Apr 13, 2019 at 14:48
• I've seen low-IF be used, to preserve the modulation phasors. Commented Apr 13, 2019 at 14:59

There are quite a few advantages.

Taking a look at a typical superhet (up to the IF):

The input signal at the RF input is small (as low as -122dBm in some narrow band voice systems I have worked on - that is about 6.3fW)

To amplify a signal at a high RF (say a few GHz) is expensive compared to doing that amplification at a lower frequency. A few dB of RF gain is usually sufficient to process the signal up to the IF stage.

The input filter bandwidth has to be such that it is less than the channel spacing (usually less than the output bandwidth) so it is easier to implement.

The image rejection filter bandwidth is set by the local oscillator frequency (at RF +/- IF), so choosing a relatively low IF means that image rejection is relatively easy to do as well.

The IF amplifier is usually where most of the signal gain is done at a relatively low cost and low complexity (compared to trying to do that at higher frequencies). The filter prevents bleed through and sets the signal bandwidth to the information bandwidth.

Another major advantage is that everything after the mixer is fixed - no adjustment necessary during normal operation which is why the dynamic signal range can be high. I have not shown the AGC (just about always present) but that is also a fixed piece of (dynamic) circuitry.

An improvement is the double superhet (2 IF stages) which I worked with decadees ago and they are still highly popular.

There are direct conversion receivers but they suffer from a number of issues, in particular dynamic range of the signal.

[Update]

In response to the comment, there are wide dynamic range direct conversion receivers (one possible source listed); these have been around for some time and are often found in SDR setups.

A purely hardware approach favours the superhet.

• Actually, some of the highest dynamic range receivers in existence are direct conversion Commented Apr 13, 2019 at 15:33
• +1 Good answer, but one point to add: Most receivers/transmitter/transceivers do not have a single frequency, but an entire range. It is easier to tune the LF part of the hardware to a single frequency, but the characatistics of the HF signal have to be maintained, which is why a carrier frequency of 0 Hz cannot be chosen directly (obviously, this would cut the LSB in most use cases). Hence, the superhet circuit converts between a tunable HF range and a lower fixed frequency; the LF part is then optimised for the fixed lower freuqency. Commented Apr 13, 2019 at 16:16

Note that this answer is skewed toward analog radio reception. The rules are different for software-defined radios, and for digital services.

The biggest drawback to direct conversion is sideband suppression. If you use a single mixer, a signal at $$\f_c + f_s\$$ is indistinguishable from a signal at $$\f_c - f_s\$$, where $$\f_c\$$ is the carrier and $$\f_s\$$ is the signal frequency. You can mitigate this problem a lot with quadrature downconversion, but while it's enough for digital data reception, it's not good enough for analog (i.e., SSB or AM).

The second biggest drawback is the amount of gain you need at one frequency. Receivers need a lot of gain, and mixers are usually the limiting factor in high dynamic range. So you'd like to put all of your gain after the first mixer, which tends to be the one that sees the most interfering signals. For a sensitive SSB receiver this gain can be in excess of 120dB. It is difficult to impossible to get that much gain without oscillating. In fact, if you look in the amateur radio literature you'll see that most direct conversion receivers have headphones -- there's a reason for this.

Other problems are LO bleed-through and the difficulty of achieving a low-noise, low-impedance amplifier at audio frequencies.

So all of these drawbacks leave you with a distinct advantage to putting at least some of your filtering and gain before the final conversion to audio -- once you assume that such a step is necessary, I think the reasons that confuse you will suddenly make sense.

• Actually, you can do sideband suppression using quadrature techniques, on TX and RX -- there are DC designs in the ham world that do just that Commented Apr 14, 2019 at 3:55
• @ThreePhaseEel yes. I addressed that: "You can mitigate this problem a lot with quadrature downconversion, but while it's enough for digital data reception, it's not good enough for analog". It can be used, and it is, but getting more than 40dB or so of opposite sideband suppression with analog circuitry requires heroic measures. The fact that only a small percentage of commercial receivers ever used it is a strong indication of the cost/benefit tradeoff. Commented Apr 14, 2019 at 14:24
• What's the "reason for this"? Commented Apr 14, 2019 at 20:58
• @DavidCullen I think you're asking about sideband suppression in direct-conversion receivers. It's because the sideband you don't want is suppressed by subtracting out a pair of signals that go through separate channels, that must be matched for gain, and which must have phase shifts 90 degrees apart. That is difficult to achieve, and subject to component variations -- so you end up with less suppression than you'd like. Commented Apr 14, 2019 at 21:20

Yes, direct-conversion receivers exist, but they require special care, especially with certain kinds of modulation.

For example, with SSB modulation, in order to reject the unwanted sideband, your baseband demodulator must be able to distinguish between "positive frequency" and "negative frequency". This is not trivial, and is only really practical using DSP.

Similarly, if you translate the center frequency of an FM or PM signal down to 0 Hz, again you need to distinguish between positive and negative frequencies in order to demodulate it properly.

Even with AM or DSB signals, where the sidebands contain identical information, unless your conversion is perfect (your LO precisely matches the received carrier frequency), you get distortion — or a constant tone from the AM carrier.

Superheterodyne was invented in the days when demodulation was done by simple envelope detectors, which didn't have any frequency discrimination at all, let alone distinguishing between positive and negative frequencies. All selectivity has to be applied before demodulation, which gets back to the reasons you found — you either need filters that can track the carrier frequency with constant bandwidth, or you convert to a fixed intermediate frequency where you can use fixed filters.

Direct conversion is conceptually simple, but requires considerable engineering to do it right. Besides Dave's & Tim's answers, there is a subtle pernicious potential problem with direct conversion...

Most mixers (even doubly-balanced ones) leak local oscillator power to both RF port and IF port. Power leaking backwards through to the RF port to the antenna can cause problems. Even with excellent mixer balance, re-radiated Local Oscillator energy is often much larger than incoming signals.
Re-radiated L.O. can be modified (phase, amplitude) and re-enter the receiving antenna, back into the mixer. Mixer balance can be adversely affected. As well, a baseband signal is generated at the mixer's output port that can foul the desired baseband signal.

Some motion sensors, "bug" detectors take advantage of this process, where the "spurious" baseband signal is the desired result, indicating motion for example.

A linear receiver is most often desired, in which case you don't want local oscillator energy to find its way back to the signal source. Having excellent mixer balance is a start, adding a strong RF preamp with good output-to-input isolation helps too. All difficult as frequency goes up.

No one else has mentioned the problem of frequency offset in the mixer stages. The RF mixer does its best to match the frequency of the carrier, but there will still be a difference which will result in the IF having an offset from the desired frequency. An IF stage with a non-zero center frequency allows a PLL to be used to track the IF signal and remove most of the frequency offset.

It is usually easier and cheaper to make the best PLL at IF frequencies rather than RF frequencies.

You can use two mixers to convert the received signal to I and Q, but that would require two expensive RF frequency PLLs that also have to have a fixed 90 degree phase difference. This is also something that is a lot easier and cheaper to do at IF frequencies.

The problem is mixer LO rejection, and close to carrier LO phase noise. 20dB is trivial out of the box. 40dB is straightforward with care. 60dB is just possible with heroics and dynamic adjustment. 80dB can be done in the lab, for a short while, at spot frequencies, if you're lucky, and the temperature doesn't change. You need >120dB to get a direct conversion receiver to work as well as a superhet.

Unless you're using OFDM. There you have dozens of carriers, and not all are used for data. There are some for pilots, some for guard bands, and in some advanced systems there are some that are only used to reduce the peak RF voltage of the ensemble to ease transmitter linearity issues. So leaving a few carriers with no data on them at and around the channel centre (DC if we're talking IQ basebands, the LO for a direct conversion receiver) is no huge inefficiency.

I found a scholarly article that seems pretty good at https://www.arrl.org/files/file/Technology/tis/info/pdf/9208019.pdf I have never made a direct conversion receiver but I have all the parts. My guess is that it will make sounds similar to a superheterodyne that has its beat frequency oscillator on. I would make any kind of HF receiver with an RF amplifier as the first stage because I already made several for my preferred bands and they would get rid of most out of band signals. I have audio amplifiers available from computer accessories so that is as good as done. I have made audio amplifiers from discrete parts and I found out that three stages of amplification (which means three transistors) is about all that I can do. With three stages (about 10 dB each) the noise of the first transistor can be heard. Adding any more stages just makes the noise louder so this appears to me to be a fundamental limit. Getting up to 100 dB gain with direct conversion seems impossible to me since 10 db for an RF amplifier and 10 dB for a converter is all I can imagine. Altogether that is only 50 dB. Superheterodyning with 30 or more dB of IF gain added to the 50 gets much closer to the 100 or more dB ideal.