Part 1 showed that a single tuned circuit and a diode cannot deliver useful selectivity or sensitivity. A heterodyne receiver solves this by mixing the signal with a local oscillator (LO) to reach a fixed intermediate frequency, where sharp filters and stable gain are practical. A homodyne receiver sets the LO at the carrier itself, so the signal lands directly at baseband. This part compares these receiver architectures with the tuned radio frequency receiver that came before them, then looks at the two blocks that make frequency translation work: the LO and the mixer.
The Tuned Radio Frequency Receiver
The obvious way to improve a detector receiver is to add RF amplifiers and more tuned circuits ahead of the diode. The result is the tuned radio frequency (TRF) receiver, where several filter and amplifier stages are all tuned to the incoming signal before detection. Because no mixer is present, the TRF receiver has no image response and no LO to radiate. It also has two serious faults.
First, every stage has to track every other stage across the band, which was hard enough with ganged capacitors and is no easier now. Second, bandwidth changes with tuning. A tuned stage has a bandwidth of f/Q, and Q stays roughly constant while f moves. With Q = 60, the bandwidth is 9.0 kHz at 540 kHz and 26.7 kHz at 1600 kHz, as Figure 2 shows. The same set is sharp at one end of the band and broad at the other, and gain varies too, which makes the chain harder to keep stable.
Both faults come from filtering at a frequency that moves. Moving the signal to a frequency that does not move removes them, and that step requires a mixer.
Heterodyne Mixing
A mixer multiplies two signals. Multiplying cos(2πfRFt) by cos(2πfLOt) gives two new components at fRF + fLO and |fRF − fLO|, and any modulation on the RF signal is carried over to both. A filter then keeps the one that is wanted, usually the difference in a receiver. The word heterodyne refers to this process of beating together two different frequencies. Homodyne means the two frequencies are the same.
The Homodyne (Direct-Conversion) Receiver
In a homodyne, or zero-IF, receiver the LO sits at the carrier frequency, so the difference product falls at DC. One fixed RF filter handles band selection and low-pass filters at baseband handle channel selection, which suits integration on a chip because no high-Q IF filter is needed. There is no separate image band to reject.
A single real mixer is not enough. With the LO on the carrier, the upper and lower sidebands fold on top of each other at baseband, and for a signal whose two sidebands differ, the information is lost. Two mixers driven 90 degrees apart solve this by producing in-phase (I) and quadrature (Q) outputs, which together preserve amplitude and phase. The I/Q demodulator is therefore the heart of the homodyne receiver.
The penalties all appear near DC. LO energy that leaks to the mixer RF port mixes with the LO and creates a DC offset. Strong signals generate second-order products at baseband, flicker noise is highest at low frequencies, and I/Q mismatch leaves a residual image. Modern processes and calibration have made these manageable, and most handset and Wi-Fi receivers now use direct conversion. Part 12 covers each impairment in detail.
The low-IF receiver
A low-IF receiver moves the signal just away from DC, often by a channel width or two, and so escapes the DC offset and much of the flicker noise. The price is an image that now falls on a nearby channel, close enough that no RF filter can remove it. Quadrature processing must cancel it instead. A receiver with 1% gain error and 1 degree of phase error between I and Q reaches about 40 dB of image rejection, which is why low-IF designs suit narrowband, low-power radios more than systems facing strong neighbouring signals.
The Superheterodyne Receiver
The superheterodyne places a tunable LO a fixed distance from the wanted signal, so every channel lands on the same intermediate frequency (IF). All the sharp filtering and most of the gain can then be designed for that one frequency. Superheterodyne receivers often use two or more conversions, with a high first IF to ease image rejection and a lower second IF to ease channel filtering. Even with more parts than a homodyne, this is still the architecture chosen when selectivity and dynamic range matter most, as in test instruments and many microwave links.
Worked example: an FM receiver frequency plan
Take an FM broadcast receiver with the standard 10.7 MHz IF, tuned to 98.1 MHz with high-side injection. The LO is 98.1 + 10.7 = 108.8 MHz. A signal at 108.8 + 10.7 = 119.5 MHz also differs from the LO by 10.7 MHz, so it lands on the IF as well. That frequency is the image, and once it reaches the mixer it cannot be separated from the wanted signal. Figure 3 shows the plan.
To tune 88 to 108 MHz, the LO must cover 98.7 to 118.7 MHz, a ratio of 1.203. Low-side injection would need 77.3 to 97.3 MHz, a ratio of 1.259, and would put its image inside other services below the band. High-side injection also keeps the image above the FM band.
A preselector must attenuate the image before the mixer. For a single tuned circuit, the normalized detuning at the image is x = Q(f/f0 − f0/f). With Q = 40 at 98.1 MHz, x = 15.9 at 119.5 MHz, giving 24.0 dB of rejection. Two such circuits, buffered from each other, give about 48.1 dB. Raising the IF moves the image further away and reduces the filter order, at the cost of a harder IF stage. Image frequency rejection and frequency planning take that trade further.
Two conversions ease the same trade. A first IF of a few hundred megahertz puts the image far from the wanted signal, so a simple RF filter removes it. A second mixer then brings the signal to a low IF, where a sharp and inexpensive filter selects the channel. Each extra conversion adds an LO, a mixer and more spurious products to manage, so the number of conversions is kept as low as the selectivity target allows.
Homodyne and heterodyne receivers compared
The table summarizes the receiver architecture choices. In practice, the choice follows from the system: a phone that must cover dozens of bands cannot carry a fixed IF filter for each, so it uses direct conversion, while a spectrum analyzer that must reject images and spurs at all costs still uses several conversions.
| Property | Superheterodyne | Homodyne (zero-IF) | Low-IF |
|---|---|---|---|
| Channel filter | Fixed IF filter | Baseband low-pass | Complex or digital filter |
| Image | Rejected by RF filter and IF choice | No separate image band | Rejected by I/Q balance |
| Main impairments | Image, spurs, filter size | DC offset, flicker, IP2, I/Q errors | Nearby image, I/Q errors |
| Integration | Moderate, external filters | High | High |
| Typical use | Instruments, microwave links | Handsets, Wi-Fi | Short-range, low-power radios |
Generating the Local Oscillator
Nearly all modern LOs are synthesized. A voltage-controlled oscillator is phase-locked to a stable reference through an integer-N or fractional-N divider, and direct digital synthesis is used where very fine steps or fast hopping are required. Part 10 develops these techniques. From the receiver point of view, an LO has four requirements:
- Tuning range and step size. The LO must reach every channel, in steps no larger than the channel raster.
- Phase noise. It must be low at the offsets where neighbouring signals can appear, not only at one convenient offset on the data sheet.
- Drive level. The mixer needs a specified LO power to reach its rated conversion loss and linearity, and a buffer amplifier may be needed.
- Power consumption. The synthesizer can be a large part of the receiver supply budget in portable equipment.
Worked example: reciprocal mixing
LO phase noise does not stay on the LO. When a strong blocker enters the mixer, the LO noise skirt is transferred onto the blocker and spills into the wanted channel. For a blocker Pb, phase noise L in dBc/Hz at the blocker offset, and channel bandwidth B, the in-channel noise is Pb + L + 10 log10B.
With a −30 dBm blocker, −125 dBc/Hz at its offset and a 200 kHz channel, the result is −30 − 125 + 53.0 = −102.0 dBm. The receiver thermal floor, for a 6 dB noise figure in the same bandwidth, is −115.0 dBm, so the LO adds noise 13.0 dB above the floor and the receiver is desensitized by about that amount. Holding the product 10 dB below the floor requires −148 dBc/Hz at that offset. A clean synthesizer can matter as much as a good LNA when strong neighbours are present.
RF Mixers
A mixer needs a nonlinear or switched element, and the choice of device and topology sets its loss, linearity and isolation. Mixer types are covered at more length in the RF mixers section. The main families are listed below.
| Mixer | Structure | Strengths | Weaknesses |
|---|---|---|---|
| Single-ended | One diode with filters | Cheap, works into millimeter-wave bands | Poor isolation, many spurs |
| Single-balanced | Two diodes and a balun | Cancels LO noise at the IF port | Partial spur rejection |
| Double-balanced | Four-diode ring or FET quad | Isolation on all ports, strong linearity | More LO power needed |
| Image-reject | Two mixers in quadrature | Cancels the image without an RF filter | Limited by balance |
| Gilbert cell | Active transistor multiplier | Conversion gain, low LO power, easy to integrate | More noise, lower linearity than a passive quad |
| Passive CMOS | Switched MOS transistors | Strong linearity on modern processes, low flicker | Needs sharp, low-noise LO drivers |
A passive mixer has a theoretical minimum loss. An ideal commutating mixer multiplies the signal by a ±1 square wave, whose fundamental has amplitude 4/π. The difference product carries half of that, giving 2/π in voltage, or −3.92 dB. Real diode rings add transformer and diode losses and typically reach 5 to 7 dB. Loss in a passive mixer behaves as noise, so its noise figure is close to its conversion loss, which is one reason an LNA sits ahead of it. Active mixers give gain instead and need less LO drive, but they compress earlier.
Mixers also produce far more than the sum and difference. Every harmonic of each input can combine with every harmonic of the other, and those combinations set the spurious performance of the whole radio.
Next in the Series
Part 3, Mixer Intermodulation Products and Intercept Points, follows that thread: how two-tone products arise, why third-order terms land inside the channel, and how intercept points predict their level and cascade through a receiver.