For the PDF version of this article, click here.

After years of development, EDGE services are now being used on GSM networks worldwide. Analysts indicate more than 100 million EDGE handsets were shipped in 2005, and that number is expected to grow to 350 million by 2008. Eventually, built-in EDGE services are expected to become a required feature for all mid- and high-end handsets. Many network service operators have announced plans to include EDGE as a complement to the 3G network, operating as a fallback data service in dense urban areas or rural areas where full W-CDMA service is not available.

EDGE is a low-cost upgrade to GSM packet data networks (GPRS), boosting data by nearly a factor of three to 120 kbps to 160 kbps. A recent survey by the Global Mobile Suppliers Association (GSA) confirms that 185 operators in 96 countries are deploying GSM/EDGE, compared with 128 networks at the end of 2004 (44% growth).

Unlike the increasingly crowded GSM handset market, EDGE is dominated by a few top-tier manufacturers. These EDGE handset manufacturers have experienced numerous manufacturing delays, due to the production challenges of existing EDGE radio designs. Radio architectures based on polar loop or polar modulation have been affected due to complicated EDGE radio calibration requirements.

To enable high-volume production of EDGE handsets, manufacturers are demanding EDGE alternatives that can provide handset performance and yield equivalent to GSM handsets. Components need to be multisourced, just like in the GSM handset market today. Achieving a handset receiver sensitivity of -110 dBm or better is becoming the industry standard.

To deliver these challenging requirements, some handset manufacturers are opting for a unique dual-transmitter architecture that preserves the best-in-class performance for GSM with a simple, one-step calibration procedure for EDGE. Unlike conventional linear transmitter architectures, no transmit SAW filter or balun is required. This architecture can be designed to be compatible with power amplifiers (PAs) and basebands from multiple vendors in the market, allowing components to be multisourced.

Figure 1 shows an example of a linear transmitter architecture compared to a polar transmitter architecture. In the polar loop transmitter, the signal is applied to the power amplifier through separate amplitude and phase feedback pathways. Polar modulation is a variant of polar loop that operates without feedback from the PA. In both cases, the amplitude pathway contains circuitry whose delay must be matched very closely to the delay through the phase pathway to avoid serious performance degradation. In a production environment, delays must be matched to account for variations in PCB manufacturing, supply voltage, frequency, output power, and temperature, creating a difficult design and manufacturing challenge. Under conditions of high PA gain, the polar loop may become unstable, causing damage to the PA or dropped calls. External loop filters complicate the production calibration by allowing unwanted parasitic coupling to the PCB, PA and other sources of interference and noise.

Figure 2 shows a simulated 8 PSK modulation spectrum for various values of delay mismatch. The 3GPP standard requires a maximum spectral power of -54 dBc at a frequency offset of 400 kHz. For high-volume production, handset manufacturers typically require that at least -60 dBc be produced from the transceiver. With a delay mismatch of approximately 30 ns, the spectrum becomes marginally failing. To put this in perspective, the GSM network operates based on a symbol period of T = 3.7 µs. The tolerable amount of delay mismatch is only a fraction of a symbol period, Δt < 30 ns = 0.008T, necessitating very high precision in the delay-matching calibration procedure for polar architectures, as well as extensive coverage requirements to account for all sources of variation in the transmitter chain.

There are alternatives to simplify calibration. As shown in Figure 1, a unique dual-transmitter architecture approach allows the offset phase-locked loop (OPLL) to be modulated by the baseband I and Q signals. In this mode, the DVGA and upconversion mixer are by-passed to add minimal noise contribution, eliminating the need for any transmit SAW filter. In 8 PSK mode, the OPLL is unmodulated and acts as a local oscillator that upconverts the baseband I and Q signals.

This architecture completely avoids the delay mismatch calibration required by polar architectures by transmitting the signal amplitude and phase together. For calibration, it requires only a simple one-step procedure for EDGE to measure output power. A similar procedure is standard practice for GSM handsets today. By comparison, polar architectures require a minimum of four calibration procedures to account for IQ predistortion, IQ dc offsets, delay matching, and output power. Table 1 summarizes the calibration requirements.

One of the stated advantages of polar architectures is high-power-added efficiency (PAE) during 8 PSK transmission. However, designers are concerned with handset talk time, a key measure of performance, which depends not just on PAE, but on the average current drawn from the battery for all key blocks in the radio including the PA, transceiver and baseband. An expression for talk time may be written as

where K is a constant that depends on the battery characteristics, and I_A is the average radio current. To obtain the longest talk time, the average radio current must be minimized.

For EDGE handsets, the average radio current, I_A, depends on the usage of the transmitter in GMSK and 8 PSK modes. Even for high-end EDGE handsets and PDAs, the GMSK mode will dominate transmitter usage. Speech calls, as well as low data rate EDGE packet data transfers are all performed in GMSK mode. The 8 PSK mode is reserved only for very high-speed data transfers, which require close proximity to the cellular base station.

Table 2 shows the average radio current calculated assuming a 90% usage in GMSK mode and 10% usage in 8 PSK mode. Despite high PA current for 8 PSK, the average radio current is comparable to what is possible for linear and polar radio.

Receiver sensitivity is a key performance metric for every GSM handset. When measured in conducted mode, a receiver sensitivity of -110 dBm or better is the industry standard for GMSK signals. However, when measured with blockers present alongside the wanted signal, achieving excellent receiver sensitivity is challenging, especially for networks with EDGE services.

Figure 3 shows an example of a digital low-IF receiver architecture compared with a direct-conversion receiver architecture. Achieving adequate AM suppression is difficult for direct-conversion receivers. In a direct-conversion receiver, the blocker passes through the switch, SAW filter and LNA and can leak over to the local oscillator (LO) side of the mixer. If the leakage is excessive, the blocker mixes with itself and produces a dc offset at the output, corrupting the desired signal. For EDGE, the blockers can be modulated in amplitude and phase, preventing a simple subtraction or averaging algorithm for dc offset correction. Depending on the algorithm employed, the sensitivity may become significantly degraded.

If a digital low-IF receiver is used, it prevents dc offsets from corrupting the signal. The dc offsets are mixed away from the signal and filtered out by the digital filter. Hence, no dc offset correction algorithm is required.

An additional concern is the integration of the synthesizer loop filter, which is not usually integrated on direct-conversion transceivers. The result is that the total bill of materials (BOM) increases because precision low-noise components are needed, and a coupling mechanism is created for external noise sources at the PCB level to add to the phase noise of the local oscillator. If the phase noise is excessive, the blocker can mix with the phase noise to produce additional low-frequency distortion that corrupts the received signal. This effect is called “reciprocal mixing,” and is a significant concern in communications systems design.

After years of development, EDGE services are a reality for global GSM networks. Handset radios based on a dual-transmitter architecture and a digital low-IF receiver architecture preserve the best-in-class performance for GSM with a one-step calibration procedure for EDGE, which reduces time to market without increasing production costs. Such handsets can offer the highest performance and be compatible with PAs and basebands from leading vendors in the market while still supporting high-volume production requirements for 2006 and beyond.

Patrick N. Morgan is a marketing manager for wireless products at Silicon Laboratories responsible for the GSM/GPRS/EDGE/3G transceiver product line. Prior to Silicon Labs, Morgan was an assistant professor of Electrical Engineering at Texas A&M University where he led a research program in electromagnetics and imaging systems. Morgan holds a Ph.D. in Electrical Engineering from Stanford University, and has authored 25 technical publications and holds two U.S. patents.

RSS    Save to Del.icio.us  Digg This