All-Constellation Receiver: GNSS Location Hub for Smartphones with Galileo Support

This tri-band receiver technology, when combined with baseband search and track engines, allows true simultaneous tracking of all current L1 GNSS signals, including GPS, GLONASS, BeiDou, Galileo, Quasi-Zenith Satellite System (QZSS), and satellite-based augmentation systems (SBAS).

By Charles Norman and Andreas Warloe, Broadcom Corporation

Starting with the first commercial GPS receivers, adding support for incrementally more complex GNSS systems presents significant challenges for GNSS hardware and software developers. The latest systems, especially Galileo, were designed with the assumption that Moore’s law would provide nearly unlimited computing resources and memory over time. The expected improvements in ASIC technology have indeed occurred, but market demands have pushed the size, cost, and power consumption of GNSS chipsets down, rather than allowing capabilities to grow freely.

GNSS in cellular phones is now expected to be always-on and to add only a few dollars to the cost of a $600 smartphone. Even as customers and phone manufacturers demand GLONASS, BeiDou, and Galileo support, chipset cost is not allowed to increase significantly. Instead of, in essence, designing four separate GNSS receivers in the chip, cost and size pressures force designers to look for commonality among the signals in order to share hardware blocks and software or digital signal-processing algorithms.

GNSS L1 Signal Down-Conversion

Commercial L1 GNSS signals span a 50 MHz range. It is getting harder for a single antenna to cover the entire bandwidth, but it is possible. The radio input contains three frequency bands of interest, spanning a total of 15 MHz:

• BeiDou, at 1561 MHz, is at the low end;
• GPS, Galileo, satellite-based augmentation systems (SBAS), and Japan’s Quasi-Zenith Satellite System (QZSS), at 1575 MHz, are in the middle; and
• GLONASS, at 1602 MHz, is at the top.

The radio process in the new tri-band receiver described here first amplifies the signal using a low-noise amplifier (LNA) to keep the system noise figure as low as possible. Then it downconverts to an intermediate frequency (IF) and filters the three bands into separate channels. The three bands are then digitized and sampled at the lowest possible sample rate. The sampled bands can be filtered digitally to remove blockers and downconverted to baseband. The baseband samples are buffered by constellations to allow parallel access for searching or tracking on each visible satellite.

All satellites in a code-division multiple access (CDMA) constellation can share baseband buffers, but the frequency-division multiple access (FDMA) constellation, GLONASS, uses a separate buffer for each satellite. This is because the memory and power required to store each satellite in use is less than storing the entire FDMA bandwidth.

Signal Similarities and Differences

All GNSS satellite signals use binary phase-shift keying (BPSK) modulation. The biphase modulation is generated from a high rate pseudorandom noise (PRN) code that is exclusive-ORed with a low-rate data stream.

The PRN code for all constellations except Galileo is generated from linear feedback shift registers (LFSRs). Galileo’s PRN code is a memory code with a bit-offset carrier BOC(1,1)/BOC(6,1) modulation. All constellations except GLONASS are CDMA. Each satellite in a CDMA constellation is at the same frequency but has a unique PRN code. GLONASS is FDMA. Each visible GLONASS satellite has a unique frequency, but all use the same PRN code.

L1 GNSS constellations use four different code lengths: 511, 1023, 2046, and 4092. The code length has a large impact on the power required to detect a signal. Data modulation is different on each constellation. BeiDou data is exclusive-ORed with a secondary code. Galileo has a secondary code-only channel. The highest data or secondary code rate is 1 kHz on BeiDou, and the lowest is 50 Hz on GPS. Table 1 shows a detailed chart with the main signal parameters for all L1 GNSS signals.

Table 1. Parameters for all L1 GNSS signals.

Radio Overview

The radio processing starts with a LNA, which utilizes a 72-nanometer negative metal oxide semiconductor transistor in a cascade configuration, with deliberate capacitive feedback and inductive source degeneration to achieve an excellent noise figure (~1.5 dB system noise figure) while maintaining a good input match. Two external matching components are required to achieve an optimal input match.

Following the LNA is an in-phase/quadrature ring mixer switched-capacitor mixer. With this style of mixer, the LNA output is only connected to one mixer output at a time and, thus, the optimal noise figure is obtained. By switching the output of the LNA from the I+ output and then later to the I– output, a 2:1 voltage gain is achieved. This improves noise figure and eases the noise requirements of the IF amplifier following the mixer, thus reducing power consumption.

The local oscillator for the mixer is derived from a low-power, low phase-noise, phase-locked loop. It has many adjustments, so the circuit can be adapted to a wide variety of reference frequencies and system requirements. It employs a ΔΣ modulator in the feedback loop, allowing for very fine frequency-control resolution.

The complex IF output from the mixer is amplified by a transimpedance section followed by three parallel amplifier/filter/attenuator sections, one for GPS/Galileo/SBAS/QZSS, one for GLONASS, and one for BeiDou. The transimpedance section’s response is close to a simple pole but with a small amount of peaking. Each of the remaining sections is built with a single complex band-pass/band-notch section, followed by real poles and zeroes. Using real poles and zeroes considerably reduces the noise and bandwidth requirements of the amplifiers. The net effect is that the power consumption of the overall IF amplifier section is substantially reduced.

There are three parallel ΔΣ analog-digital converters (ADCs), one for each of the three IF sections. The ΔΣ ADC is a continuous-time, second-order, one-bit ΔΣ ADC, running at a sample rate of 395.75 Msps. The ΔΣ ADC comprises two operational amplifiers, two digital analog converters, and a quantizer. The ΔΣ ADCs are designed in such a way that the quantization noise is lowest not at zero frequency offset (DC), but at the offset frequency of the GNSS signal. The A/D samples are filtered with a third-order cascaded integrator-comb subsampled at 99.44 mega-samples per second. Additional finite impulse response (FIR) filters and subsampling to 33.1 MHz complete the sampling. The combined ΔΣ ADC and digital filtering provide more than 50 dB of dynamic range.

Digital processing at 33.1 MHz includes several filters that remove interference sources from the received radio signal and automatic gain control logic that adjusts the gain of the IF amplifiers to give an optimal signal level. A configurable 20-tap FIR filter is provided for each sample section and can be configured to remove wideband blockers. In addition, each section has eight narrowband, single-pole infinite impulse response filters for removing narrowband blockers.

Figure 1. Radio overview diagram.

Separate Search and Track Blocks

Separate search and track sections are employed to compute correlations between the three sample streams and multiple reference hypotheses. The three sample streams are buffered in memory to allow the search and track sections to process multiple correlations in parallel. Search employs a prime factor fast Fourier transform with a selectable size (1023, 2046, or 4092).

Search correlations are computed by first removing a hypothesis Doppler from a buffered set of samples and then combining a selectable number of code epochs. The filtered samples are translated to the frequency domain, multiplied by the frequency-domain representation of the desired PRN code, and finally translated back to the time domain. This process creates a coherent correlation vector for the entire code. The coherent correlation vector is non-coherently accumulated until the signal-to-noise ratio of the peak exceeds a detection threshold.

Track correlations are computed in the time domain by multiplying a multichip reference code by a set of buffered samples. Typically, the reference code is linearly delayed for N correlations to produce an N-sample coherent correlation vector. The correlation vectors are buffered to allow multiple filters to be processed in parallel. A coprocessor is used to run the filters. The outputs from the coprocessor provide estimates of code phase, Doppler, acceleration, data synchronization, data bits, signal power, and more.

All the buffering and multiple processing sections allow for multiple hypotheses to be tested in parallel. For example, on a tunnel entry, the attenuated signal can continue to be tracked while the search section tries to detect the full-power signal.

Secondary Code Resolution. Several constellations have secondary codes that limit the length of the coherent integration unless the code can be wiped. GLONASS has a 100-Hz Manchester code, BeiDou has a 1-kHz secondary code, and the Galileo Pilot has a 250-Hz secondary code. After the time accuracy drops below 1 millisecond, all of the secondary codes can be wiped in both search and track, so the coherent period can be optimized to maximize sensitivity and minimize measurement error. On a cold start, when time is unknown, it is best to first try to detect with coherent correlations less than the secondary code chip period.

When a signal is detected, the receiver either goes into track and computes correlations with longer coherent periods for multiple time hypotheses or continues in search with a longer coherence period and multiple time hypotheses. The search and track sections allow for either of these choices. For constellations like Galileo, the best choice is to remain in search. For others like BeiDou, it is best to move to track.

Benefits of Multi-GNSS Receivers

The ability to track all L1 constellations means that even in difficult environments, there are a sufficient number of satellites to produce a navigation solution. As can be seen from field-test results, not only are more satellites tracked, but more satellites with strong signals are tracked. The measurement errors of satellites received with strong signals will be smaller, leading to very low bit-error rates and allowing for a faster ephemeris collection. Field test results confirm that a receiver with BeiDou support achieves faster and more accurate fixes than a receiver without BeiDou support (see Figure 2).

Figure 2. A receiver with BeiDou support achieves faster and more accurate fixes than a receiver without BeiDou support.

In addition to speed and accuracy improvements, more constellations provide a higher reliability. Recently, an upload error in the GLONASS constellation caused otherwise healthy satellites to report orbit errors of several kilometers. GPS/GLONASS-only systems could not completely isolate the faulty satellites. In difficult environments, there are not enough good satellites to isolate the faulty ones. With the addition of BeiDou, the faulty satellites were correctly isolated (Figure 3).

Figure 3. (Top) Seoul, South Korea, third-party GPS/GLONASS-only receiver;
(bottom) Broadcom GPS/GLONASS/BeiDou receiver enables isolation of faults.

Each constellation adds unique improvements. Narrowing the correlation triangle allows for improved multipath rejection and more accurate pseudorange measurements (Figure 4).

Figure 4. Narrower correlation triangle.

GLONASS, with the slowest code rate, has the broadest correlation triangle. BeiDou, with the highest code rate, has a correlation triangle that is narrower than GPS. The BOC code on Galileo gives the narrowest correlation triangle. Field test results confirm the improved measurements (Figure 5).

Figure 5. Left, mean pseudorange measurement error, urban streets; Right, rural freeway.

GLONASS, the only FDMA constellation, has the least cross-correlation. GPS uses Gold codes to keep the cross-correlations between any of its satellites at a minimum. BeiDou and Galileo have lengthened their codes and added a secondary code to reduce cross-correlations.

Conclusion

Taking advantage of similarities in the L1 GNSS constellations together with careful design choices to minimize size and current consumption has enabled the creation of commercial GNSS system-on-chips that support all current GNSS L1 systems and meet the cost, size, and power requirements of cellular phones. The addition of new constellations like BeiDou and Galileo has significantly improved speed, performance, and reliability.

Acknowledgments

Javier de Salas, Frank van Diggelen, and John Hutson, all of Broadcom.

Manufacturer

The BCM4774 single-chip GNSS location hub for smartphones with Galileo support was designed by Broadcom Corporation.

Charles Norman is a technical director in the GNSS group at Broadcom Corporation. Previously, he worked on GNSS systems at Magnavox, Interstate, SIRF, and RFMD. He holds 39 issued patents on GNSS systems and has an M.A. in mathematics from the University of California-Los Angeles.

Andreas Warloe is a senior technical director in the GNSS group at Broadcom Corporation. He previously worked on GNSS receivers at Magellan, Leica Geosystems, IBM, and RFMD. He holds an M.S. in electrical engineering from the University of Southern California.