RF System Architecture

A radio has two jobs: receive a tiny, faint signal from the air and turn it into data your phone/computer can use — and transmit your data as a powerful, clean signal back into the air. Neither is trivial.

The received signal might be −90 dBm (one billionth of a milliwatt — almost nothing). The noise from the environment is at −100 dBm. A strong interference signal from a nearby transmitter might be −30 dBm simultaneously — that's 70 dB stronger (10 million times more powerful) than the signal you're trying to receive. How do you amplify the weak signal without also amplifying the interference by 10 million? That's the core challenge that radio architectures are designed to solve.

The receive chain goes: Antenna → Filter → LNA → Mixer → IF Filter → VGA → ADC → Digital processing. The transmit chain goes the other way: Baseband → DAC → Mixer → Filter → PA → Antenna. The key question is: how do you arrange these blocks to get the best performance? That's what the different architectures decide.

Invented by Edwin Armstrong in 1918, the superheterodyne (or "superhet") design has been the dominant radio architecture for over 100 years. Every AM/FM radio, every traditional TV, every early mobile phone used it. The idea is elegant: move the incoming signal to a fixed intermediate frequency (IF), then do all the hard filtering and amplification at that fixed frequency.

Think of it like this: instead of building a different equaliser for every possible incoming frequency, you shift everything to a standard frequency first — like always converting foreign currency to dollars before spending it, no matter which country you're in.

The secret weapon of the superhet is the IF filter. At IF (say, 70 MHz), you can use a crystal or SAW filter that has extraordinary selectivity — it can reject signals just 200 kHz away from your desired channel with 60 dB of attenuation. If you tried to build that same filter at 900 MHz (the RF frequency), it would be nearly impossible — the fractional bandwidth would be so small that the filter would be either impractical or physically huge.

By converting everything to a fixed IF first, the radio designer has one excellent filter they can use for all possible incoming channels. Just change the LO frequency to tune to a different channel — the IF filter stays the same. It's like having a universal adapter: whatever plug the signal comes in as, you convert it to a standard shape and then process it.

The mixer's job is to shift f_RF to f_IF. It does this by multiplying by the LO frequency. But multiplication creates two outputs: f_RF − f_LO = f_IF AND f_LO − f_image = f_IF. There's a frequency on the other side of the LO that also mixes down to the same IF! This is the image frequency.

The solution is the image reject filter — a bandpass filter placed before the mixer that passes f_RF but blocks f_image. This filter only needs to reject the image frequency, not a nearby interferer, so it's much easier to build than the channel-select IF filter. Higher IF frequency = wider separation between f_RF and f_image = easier image filter.

Direct conversion (also called zero-IF or homodyne) takes a different approach: skip the IF stage entirely and convert the signal straight to baseband (0 Hz). The LO frequency is set exactly equal to the carrier frequency. The incoming RF signal mixes with the LO and comes out at the difference frequency — which is the baseband signal itself.

When you set f_LO = f_RF, there's a problem: some of the LO signal leaks into the LNA input, mixes with itself, and creates a large DC offset right at 0 Hz — the exact frequency your signal is sitting at. This DC offset can easily be 1000× larger than your desired signal, swamping the ADC and making it impossible to recover the signal.

Modern chips solve this with DC offset calibration — measuring the DC at startup and subtracting it digitally. But bursts of interference can cause the DC offset to change, and very slow data rates (where the signal spectrum extends down to nearly 0 Hz) are still problematic. This is why direct conversion was considered impractical for decades but is now the standard in smartphone chips — modern calibration circuits are that good.

Low-IF architecture is a compromise: convert the signal to a low but non-zero IF (e.g. one channel width away, typically 100 kHz to 2 MHz). This avoids the DC offset problem of zero-IF (the signal is no longer at DC), while keeping the IF low enough that the filter can be integrated on chip.

The image problem returns, but since the IF is low (e.g. 1 MHz), the image is only 2 MHz away from the desired signal — you must handle it digitally using a complex (I+Q) signal and a polyphase filter that rejects the image in the digital domain. Low-IF is used in Bluetooth and many IoT chips.

Let's trace a signal from the antenna all the way to the ADC and see exactly what happens to the signal level and the noise at each stage. This is the "cascaded analysis" that every RF systems engineer does for every new design.

Every RF engineer learns this rule early: put the gain first, put the filtering first. Specifically:

Rule 1 — LNA first, gain first: The first element in the receive chain dominates the system noise figure. If you put a 3 dB attenuator before the LNA, you've added 3 dB to your system noise figure — which halves your range. The LNA must be as close to the antenna as possible.

Rule 2 — Filter early: Strong out-of-band signals (blockers) eat into the dynamic range of every amplifier they reach. Filter them out before they can cause intermodulation products or desensitise the receiver.

Rule 3 — Most gain at IF/baseband: Gain at IF is cheaper, lower noise, and easier to control with AGC (automatic gain control). The RF stage only needs enough gain to overcome mixer noise.

The transmit chain is the receive chain in reverse, with some important differences. The signal starts as digital data, is converted to analogue (DAC), filtered, mixed up to RF frequency, amplified by the PA, filtered again to meet spectral mask requirements, and transmitted.

The TX chain differs from RX in two critical ways. First, the PA at the end is the dominant component — it must deliver the specified output power efficiently with low distortion. Second, there's usually a TX filter or duplexer after the PA to clean up harmonics and out-of-band emissions before they get to the antenna. Regulatory bodies (FCC, ETSI) set strict limits on spurious emissions — you can face large fines for broadcasting on the wrong frequencies.

A mobile phone must transmit AND receive simultaneously — you're talking and listening at the same time. This is called full duplex, and it presents a fundamental problem: your PA is transmitting at +23 dBm, and your LNA is trying to receive a signal at −90 dBm from the same antenna. That's a difference of 113 dB — a ratio of 100 trillion to one. How do you stop the PA from destroying the LNA?

Imagine you're trying to hear someone whispering at the back of a concert hall while a rock band plays in front of you. Your ears get overwhelmed — that's the blocking problem in radio. A strong interfering signal drives the LNA or mixer into compression, reducing its gain and increasing its noise, so it can no longer detect the weak desired signal.

Adjust each stage and watch how the cascade noise figure, sensitivity and dynamic range change. This is exactly how engineers design a receive chain.