RF systems and radio architectures

The TLR-1 telemetry radio has been in the field for a year. Its 900 MHz hopping link carries push-button calls and loop-detector reports from the poles of an intersection to the controller cabinet, and in the six-week trial in Madison it delivered more than three million messages without losing one. The city's traffic engineers now want more. They want every cabinet to report its status to the traffic management center without a leased telephone line, they want technicians to read diagnostics from a pole with a phone instead of opening the housing, and they want the cabinet-to-cabinet link to reach intersections that are farther apart. Daniel Fortin, PE, has asked Amara Osei to lead the design of TLR-2, and this course follows that design from the first block diagram to the compliance plan.

ELE-350 treated the TLR-1 link as a communication system: modulation, noise, error rates and the link budget. ELE-360 treated it as an electromagnetic problem: transmission lines, reflections and the antenna. This course treats it as a set of radio-frequency circuits that have to be designed, laid out, measured and certified. The questions change from what signal-to-noise ratio is needed to how the low-noise amplifier should be matched, which filter goes first, how much phase noise the synthesizer may have, and how three radios can share one small enclosure without deafening each other. Priya Natarajan, who designed the CX-1 chip in ELE-445, joins the team for the circuit work, and Leo Tremblay, on another co-op term, builds the test fixtures and runs the measurements.

What TLR-2 must do

Amara starts from the requirements. The 902–928 MHz telemetry link stays, with the same frequency-hopping GFSK format at 50 kb/s, but its front end is redesigned: the cabinet unit gets a 27 dBm power amplifier, a surface acoustic wave (SAW) filter and an external low-noise amplifier so that cabinet-to-cabinet links can span 600 m with a large margin. A cellular modem using LTE-M or NB-IoT, the two low-power versions of LTE intended for machine connections, gives each cabinet a link to the traffic management system. A Bluetooth Low Energy (BLE) radio at 2.4 GHz in every pole and cabinet unit lets a technician's phone read status, logs and settings. A pole unit has two radios; a cabinet unit has all three.

Each radio raises different design problems. The telemetry radio is Cedarline's own design and the team controls everything in it, from the matching networks to the antenna. The cellular radio will be a certified module bought from a module maker, because designing and certifying an LTE modem is far beyond a small company; the team's work is the antenna, the cable, the layout around the module and the coexistence with the telemetry receiver a few centimeters away. The BLE radio is a system-on-chip with an integrated radio, and the work is the antenna and its matching on a crowded board.

The radio-frequency signal chain

Every radio receiver performs the same few tasks. It selects the wanted band and rejects others, amplifies a signal that may be as weak as a femtowatt, converts it to a frequency where it can be filtered and digitized, and recovers the information. Every transmitter does the reverse: it generates a modulated signal at a low frequency or in digital form, converts it to the carrier frequency, amplifies it to the required power and filters out everything the rules do not allow it to radiate. The way these tasks are arranged is the radio's architecture, and the choice of architecture fixes most of the performance before any circuit is designed.

Three quantities describe a receiver chain. Gain is the ratio of output to input power, usually in decibels. Noise figure (NF) measures how much the chain degrades the signal-to-noise ratio, and sets the weakest signal that can be received. Linearity, usually stated as the third-order intercept point (IIP3) referred to the input, or the 1 dB compression point, sets the strongest signals the chain can handle before it creates distortion products that bury the wanted signal. ELE-350 introduced all three; here they become the design variables.

Architectures

The superheterodyne receiver, invented about a century ago and still widely used, converts the incoming signal to a fixed intermediate frequency (IF) by mixing it with a local oscillator (LO). Because the IF is fixed, the channel filter and most of the gain can be built once and optimized, and the receiver is tuned by changing only the LO. The price is the image: a signal on the other side of the LO, one IF away from it, also mixes down to the IF and must be removed by a filter before the mixer. A high IF makes the image easy to filter but makes the channel filter harder; a low IF does the opposite. Many superheterodyne receivers use two conversions to get both benefits.

The zero-IF, or direct-conversion, receiver sets the LO at the carrier frequency and mixes directly to baseband in two paths, in-phase (I) and quadrature (Q), whose LOs differ by 90°. There is no image in the usual sense, because the image is the signal itself, and the channel filtering is done by simple low-pass filters that integrate well on a chip. Direct conversion has its own problems. The LO leaks to the antenna port and mixes with itself, producing a DC offset that can be larger than the wanted signal; second-order distortion converts strong amplitude-varying signals to baseband; and the flicker noise of transistors is largest at low frequencies, exactly where the signal now sits. Modern transceivers deal with these by calibration, high-pass filtering and careful layout, and direct conversion is now the usual choice for wideband cellular receivers.

The low-IF receiver, used in most sub-GHz and Bluetooth transceivers, mixes to an IF of a few hundred kilohertz to a few megahertz with I and Q paths and removes the image with a complex (polyphase) filter or in digital processing. It avoids the DC-offset and flicker-noise problems of zero-IF while keeping most of its integration benefits. The image is rejected by matching the I and Q paths rather than by a filter, so the rejection is limited to about 25 to 40 dB by gain and phase mismatch, as ELE-350 noted. The software-defined radio carries this idea further: the analog-to-digital converter sits as close to the antenna as power and cost allow, and filtering, demodulation and even the choice of standard are done in software.

The transceiver chip in TLR-1 is a low-IF design, and Amara keeps the same family for TLR-2 so that the software carries over. The cellular module uses direct conversion internally, and the BLE system-on-chip uses a low-IF receiver. The team does not design any of these architectures from scratch; it designs the circuits around them, and to do that well it must understand their weaknesses. Lesson 7 returns to the mixers and oscillators inside them.

Transmitter architectures

Transmitters have their own architectures. A direct-conversion transmitter generates I and Q baseband signals, converts them to analog and mixes them up to the carrier in one step; a polar transmitter separates amplitude and phase, driving a phase-modulated oscillator and modulating the amplitude through the power amplifier's supply. For a constant-envelope signal such as GFSK there is no amplitude to modulate at all: the synthesizer itself is modulated, and the power amplifier can run in saturation, which is why the TLR-1 transceiver can produce 14 dBm with good efficiency. The cellular module transmits signals whose amplitude varies, so its amplifier must be linear, a contrast that lesson 6 develops.

Cascade analysis of a receiver line-up

The first design decision for the TLR-2 telemetry receiver is what to put between the antenna and the transceiver chip. ELE-350 showed that the TLR-1 pole receiver, with a ceramic filter in front of a chip whose noise figure is 5 dB, has a system noise figure of 7 dB, while placing a low-noise amplifier first brings it near 1 dB. Lesson 7 of ELE-350 used the Friis formula for noise; the same cascade idea applies to linearity, with a twist that matters for TLR-2.

The noise formula says that the first stage dominates the noise figure if its gain is high. The linearity formula, with powers in milliwatts and gains as ratios, says the opposite for distortion: the later stages dominate, because they see the signal after it has been amplified. An LNA with 16 dB of gain in front of a transceiver whose IIP3 is −12 dBm lowers the input-referred IIP3 of the whole chain by about 16 dB. Gain improves sensitivity and degrades the ability to handle strong signals, and the designer has to strike a balance.

import math

# Cascade analysis of TLR-2 sub-GHz receiver line-ups at 915 MHz
def db(x): return 10 * math.log10(x)
def lin(d): return 10 ** (d / 10)

def cascade(stages):
    """stages: (name, gain dB, NF dB, IIP3 dBm). Returns gain, NF, IIP3 at the input."""
    g = 1.0; f = 1.0; inv_ip3 = 0.0
    for k, (name, gdb, nf, iip3) in enumerate(stages):
        f += (lin(nf) - 1) / g if k else lin(nf) - 1
        inv_ip3 += g / (lin(iip3) * 1e-3)        # refer each IIP3 to the input
        g *= lin(gdb)
    return db(g), db(f), db(1e3 / inv_ip3) if inv_ip3 else float('inf')

bw = 67.5e3          # noise bandwidth of the 50 kb/s GFSK channel, Hz
snr_req = 10.0       # SNR needed for the packet error target, dB
lineups = {
    'A: ceramic filter + chip (TLR-1 pole)': [('filter', -2.0, 2.0, 60), ('chip', 0, 5.0, -12)],
    'B: LNA + chip':                         [('LNA', 16, 1.0, 0), ('chip', 0, 5.0, -12)],
    'C: SAW + LNA + chip (TLR-2)':           [('SAW', -1.5, 1.5, 60), ('LNA', 16, 1.0, 0), ('chip', 0, 5.0, -12)],
}
for name, st in lineups.items():
    g, nf, iip3 = cascade(st)
    floor = -174 + db(bw) + nf          # noise floor in the channel, dBm
    sens = floor + snr_req
    sfdr = 2 / 3 * (iip3 - floor)       # spurious-free dynamic range, dB
    print(f"{name}")
    print(f"   gain {g:5.1f} dB  NF {nf:4.2f} dB  IIP3 {iip3:6.1f} dBm  sensitivity {sens:6.1f} dBm  SFDR {sfdr:4.1f} dB")

A: ceramic filter + chip (TLR-1 pole)
   gain  -2.0 dB  NF 7.00 dB  IIP3  -10.0 dBm  sensitivity -108.7 dBm  SFDR 72.5 dB
B: LNA + chip
   gain  16.0 dB  NF 1.18 dB  IIP3  -28.0 dBm  sensitivity -114.5 dBm  SFDR 64.3 dB
C: SAW + LNA + chip (TLR-2)
   gain  14.5 dB  NF 2.68 dB  IIP3  -26.5 dBm  sensitivity -113.0 dBm  SFDR 64.3 dB

The program compares three line-ups. Line-up A, the TLR-1 pole receiver, has the best linearity but the worst sensitivity, −108.7 dBm. Line-up B, an LNA straight into the chip, gives the best sensitivity, −114.5 dBm, but an input IIP3 of −28 dBm: two strong signals at −50 dBm each would create third-order products near the noise floor. Line-up C, Amara's choice for the cabinet, puts a SAW filter before the LNA. It gives up 1.5 dB of sensitivity, but the SAW removes strong cellular, paging and land-mobile signals outside 902–928 MHz before they reach the LNA, where they would otherwise compress it or create distortion. The spurious-free dynamic range, two-thirds of the distance between the noise floor and the IIP3, is the same for B and C, but C's applies only to signals inside the band, which is where it matters.

Dynamic range and blocking

Sensitivity and IIP3 are not the only limits. A receiver must also tolerate a single strong signal close to the wanted one, called a blocker, without losing sensitivity. A blocker can compress the LNA, reducing its gain; it can mix with the LO's phase noise and land in the channel, as lesson 7 shows; and it can be strong enough to overload the analog-to-digital converter. Data sheets state blocking performance as the level of an interferer at given frequency offsets that causes a specified loss of sensitivity, and the TLR-2 specification includes a blocking requirement taken from the worst signals Leo measured at three busy intersections: a cellular uplink at 849 MHz from a phone in a car stopped beside the cabinet, and land-mobile radio transmitters near 460 MHz used by the city's own trucks.

Leo's survey also found a paging transmitter near 930 MHz, only 2 MHz above the band edge, at about −35 dBm at the cabinet antenna. No affordable SAW filter can reject a signal that close by more than a few decibels, so the transceiver's own selectivity must handle it. Amara adds the measured levels to the requirement list; they will drive the filter choice in lesson 4 and the coexistence work in lesson 11.

Planning the TLR-2 project

Amara divides the work by risk. The telemetry front end is mostly Cedarline's own circuitry, so its matching networks, LNA, power amplifier and harmonic filter are designed and measured first. The cellular backhaul is lower risk in the circuits, because the module is certified, but higher risk in the system: antenna placement, cable loss, coverage at each cabinet and carrier approval. The BLE link is the simplest, but its tiny antenna must work in a metal pole housing. Daniel insists that every radio-frequency design decision be backed by a measurement on a test fixture before the full board is laid out, a rule that lesson 2 makes possible by teaching how such measurements are made.