The main benefit of EGPRS is its provision of a cost-effective way to ease into next-generation services and applications. It accomplishes this by enabling increased data transmission rates (throughput), improved quality of data services (reliability, response time) and increased spectral efficiency. It thereby enables a smooth evolution of current/existing global system for mobile communications (formerly Groupe Speciale Mobile) (GSM) and general packet radio service (GPRS) networks to next-generation applications at a relatively low upgrade cost.

The EGPRS standard offers higher data transmission rates than both GPRS (2.5G) and the GSM standards. These higher data rates are achieved by introducing new coding and modulation schemes into the air interface. The new modulation used in EGPRS is 3p/8 rotated 8-phase shift keying (8-PSK).

The new 8-PSK modulation enables transmission of a bit rate three times higher than the Gaussian minimum shift keying (GMSK) used in GSM and GPRS, but at the cost of new design challenges for both RF and base band processor (BBP) subsystems. Furthermore, cost pressures have forced cellular equipment manufacturers to find more cost-effective components, including cost-effective RF subsystems. Relatively low-cost RF subsystems introduce challenging requirements for base band subsystems where the key is to maintain the highest bit rate possible. Integration became more challenging with the introduction of coupled relations between RF and base band subsystems. Generally speaking, the two most common RF architectures to be found in coming cellular handsets are direct-conversion receiver (DCR) and near-zero intermediate frequency (NZIF). This article focuses on DCR architecture.

The cellular modem signal flow
Signal flow involving a cellular communications device is described in Figure 1.

A block of information bits is input into the transmitter BBP. The transmitter BBP executes the tasks required by the specific cellular standard such as coding, interleaving, block building, modulation, filtering, etc. It then feeds two analog signals into the RF subsystem: xI(t) and xQ(t). These signals relate to the in-phase (I) and the quadrature (Q) projections of the transmitted samples in the complex base band domain. The RF portion of the transmitter then upconverts the signal to a specific carrier frequency.

The signal passes through a multipath wireless channel that yields the received signal, which may be corrupted with interference and noise. After undergoing demodulation and filtering, the analog signals, I(t) and Q(t), are then sampled from the RF portion at the analog base band (ABB) the first stage of the BBP. Finally, the digital base band (DBB) performs filtering, equalization, estimation and decoding to recover the transmitted information.

An automatic frequency correction (AFC) loop (represented by a dotted line in Figure 1) is needed for frequency synchronization. Other feedback loops exist between the BBP and the RF subsystems with many known implementations. Some of the issues involved in EGPRS DCR handset loop design are discussed in the following sections.

The difference between GPRS and EGPRS
The GPRS network overlays a packet-switched capability onto the existing circuit-switched core network without altering the air interface. EGPRS, on the other hand, upgrades the air interface while using the same GPRS packet-switched core network to offer a packet-optimized cellular network with higher data rates. It provides maximum theoretical packet data transport rates of up to 474 kb/s using the same GSM carrier bandwidth of 200 kHz.

Developers of EGPRS systems were faced with the challenge of increasing data rates in existing GSM networks without degrading system performance. To accomplish this, EGPRS is based on two enhancements to GPRS specifications: nine new modulation and coding schemes designated MCS1-MCS9 and incremental redundancy (IR). The integration of MCS9 (unity coding gain) with the 8-PSK modulation scheme presents a challenge for both RF and modem designers.

The first enhancement supplementing the GMSK modulation used by GPRS, with 3p/8 ? 8-PSK modulation, is shown in Figure 2. The black and white dots stand for the two rotated constellations (without the rotation, there would have been only one set of dots). EGPRS supports both GMSK and 8-PSK modulation schemes.

The new modulation impacts the design of modulators, amplifiers and receivers for several reasons:

It is a non-constant envelope signal.

It is used under higher signal-to-noise ratio (SNR) conditions (about 30 dB for MCS9). Under such conditions, compensating channel and RF impairments becomes critical, especially when a phase modulation (PM) signal (3p/8 ? 8-PSK) is received.

The complexity of the equalizer is greatly increased due to a larger modulation scheme alphabet and an effective increase in the channel length.

Another important evolution in the design of next-generation handsets is the numerous possibilities for RF architectures, including DCR1-4 and NZIF5, 6. The main purpose of these architectures is to provide a compact, low-cost and low-power solution. As stated earlier, the focus here is on DCR.

The integration of RF and EGPRS BBP is one of the major challenges facing physical layer (PHY) designers. Some of the evolving RF architectures trade lower cost and smaller form factor for increased levels of impairment. But at the same time, EGPRS, as a more sensitive scheme, requires improved RF performance. The solution may come from using new signal-processing hardware impairments (HWIs), mitigation algorithms and RF subsystems tuned to EGPRS modem requirements.

EGPRS DCR main HWIs
The main reason for the increase of HWIs in a DCR RF subsystem lies in the fact that the I/Q demodulator is not isolated from the RF stage, unlike the isolation existing in a superheterodyne (SH) receiver. A general DCR architecture is depicted in Figure 3.

The RF signal is filtered and amplified before it enters the I/Q demodulator, which contains a local oscillator (LO), a phase splitter, two mixers and two LPFs. The demodulator converts the signal directly from the RF carrier to base band and simultaneously splits it into its I and Q components. This makes the new architecture cost-effective and less power consuming. This happens at the cost of an increased level of HWIs.

From the DBB point of view, the four main HWIs in a typical EGPRS DCR are:

Static DC offset.
Dynamic DC offset.
Frequency offset and phase noise.
I/Q imbalance (phase and gain imbalance).
These impairments have a greater effect on the 8-PSK modem performance than they do on a GMSK modem. Note that the effect and the level of the HWIs greatly depend on the specific RF architecture. For instance, while DC offsets are a less important issue in an SH receiver, for a DCR they become highly dominant7, 8.

It is important to understand the impact of these impairments on the effective bit rate of the EGPRS handset as EGPRS delivers the promise of next-generation data applications.

Static DC offset
Static DC is used in this article to describe the addition of a constant DC to the EGPRS burst signal, as shown in Figure 4.

The major sources of a Static DC offset are:

Self-mixing of all channels in the GSM downlink band. The received signal is coupled to the local oscillator (LO) port and creates a base band replica.

Self-mixing of the LO signal.

Analog filters in the receiver (1/f noise).

A/Ds of the receiver.

DC injection in the base station side (1/f noise and D/As).

The self-mixing contributions to the DC shift can be found only in a DCR. In addition, the levels of DC that are apparent in a DCR are much higher than those commonly apparent in SH receivers.

In a DCR, static DC offset has a tremendous effect on the throughput of both GMSK and 8-PSK modems. Furthermore, high-pass filtering, which can be used for other communications systems such as wideband code division multiple access (W-CDMA), is not appropriate for GSM. However, the EGPRS BBP can compensate static DC offset, either by a joint analog and digital solution, or by a purely digital one7, 8. The PHY designer should consider the dynamic range and calibration method. This pertains to the decision of how and where to compensate this static DC compensation in the digital domain by the BBP. This is a low-cost solution, however it requires a small and limited DC shift generation by the RF and ABB subsystems.

The term ?static? implies that the DC level is constant over many EGPRS bursts. However, an efficient and low-cost digital signal processing (DSP) algorithm can be implemented to estimate the DC shift accurately. This is based on a single EGPRS burst, and requires no historical knowledge.

It is also noteworthy that static DC, produced by any of the sources mentioned, can be compensated by the BBP of the receiver, including the DC injected into the base station. This phenomenon of by-product compensation for transmitter impairments in the receiver will be discussed again in the section dealing with frequency offset and the DSP PLL.

Note that self-mixing of the received signal (the first source mentioned) yields not only static DC. When an EGPRS signal (or any other multitone signal) undergoes self-mixing in the receiver, the result is a wideband signal with a large spectral component at the low frequency band and DC. This in-band interference is hard to suppress by BBP alone.

A special case of self-mixing occurs when a constant envelope interference signal, such as GMSK, propagated through a frequency selective wireless multipath channel (the common scenario in GSM cellular communication), is self-mixed. In this common scenario, the ?constant envelope? quality is removed from the received signal or interferer by the multipath channel. The self-mixing product in this case is again a wideband signal reflected to base band, producing an interference that is hard to suppress. Self-mixing may cause further interference called dynamic DC interference.

Dynamic DC offset
The term dynamic DC refers to the self-mixing of a non-synchronized strong interference signal.

Its dynamic nature comes from an asynchronous network for which cell bursts are not aligned. When an interference burst is transmitted and self-mixed in a DCR, it alters the DC level at the BBP input. This change may be as fast as the interferer rise time (few symbols in a GSM interferer). In an asynchronous network, it may appear at any time within a desired burst duration. This scenario is depicted in Figure 5.

As for suppression of the dynamic DC, when integrating with sufficient BBP performance, a good target for RF design would be 33% of the received signal in an amplitude modulation (AM) suppression test. Compensation of higher levels is usually hard to achieve by the BBP.

Frequency offset and phase noise
The difference between the transmitter and the receiver LO is referred to as a frequency offset. Because the base station LO is taken as the reference, frequency offset cannot be eliminated in the handset. Furthermore, frequency offset has a critical effect on the 8-PSK equalizer. When its level is too high, it becomes exceptionally difficult to compensate for it using BBP alone. The error rate increases and synchronization of the mobile can be lost. For these reasons, frequency offset is an important issue ? in particular, with regard to RF and BBP integration.

To minimize frequency offset effects, both pure BBP and joint compensation loops are used. After synchronization, the input to the BBP has a residual frequency offset. According to GSM standards, frequency offset should be below 0.1 ppm (e.g., 190 Hz for a carrier of 1900 MHz). To achieve the required compensation accuracy, a preliminary synchronization is applied.

A joint process of RF and BBP executes the preliminary synchronization of a cellular terminal to the GSM network9. To maintain synchronization during the connection, the AFC loop is applied. This loop gets a digital estimation of the previous frequency offset, which is fed into a phased-locked loop (PLL) that controls the frequency of the LO.

This loop implementation is the same for both GSM and EGPRS. However, in EGPRS 8-PSK modulation, there is a need for a digital PLL to further compensate for the residual frequency offset. Residual frequency offset compensation during the equalization process is important when using 1900 MHz DCS and 1800 MHz PCS frequency bands. This is due to the higher residual offset at base band for the same ppm error. The effect of an uncompensated residual frequency offset for 8-PSK is illustrated in Figure 6. Maximum phase drift (18.7ֲ°) was calculated for 190 Hz frequency offset (0.1 ppm of 1900 MHz) and for one-half a burst length (74 samples).

Unlike frequency offset, phase noise is difficult to compensate because of its random nature. The solution is to suppress it at the RF stage to a level that is reasonable for the BBP. Phase-noise suppression is even more important when integrating EGPRS RF and BBP:

8-PSK is transmitted under conditions of high SNR (about 30 dB), in which phase noise becomes dominant (higher throughput).

As a PM signal, 8-PSK is highly sensitive to phase noise.

I/Q imbalance (phase and gain imbalance)
The I/Q imbalance consists of two components: gain and phase. The phase imbalance source is the 90ֲ° phase splitter (refer to Figure 2 ? DCR block scheme). The splitter inaccuracy is reflected to the base band as unwanted correlations between the I and the Q components of the signal (see Figure 7).

The gain imbalance (see Figure 8) is the result of two different analog paths for I and Q (filters and amplifiers).

When using an SH receiver, those impairments are quite small. With a DCR, they affect the performance for high SNR conditions only. Notice that the 8-PSK modulation will be used under high SNR conditions, making it a more important issue for EGPRS than for GSM/GPRS.

Both RF and BBP stages may achieve the I/Q imbalance mitigation for a DCR. The first mitigation method is to reduce it in RF to a low enough level (less than 3ֲ° phase imbalance and less than 0.8 dB gain imbalance). These levels of I/Q imbalance cause a degradation of throughput, but standard requirements can still be fulfilled. The second method is to use a BBP algorithm to estimate and compensate for these imbalance imperfections. This loop can be designed to be efficient, displaying low complexity and low power consumption.

Summary
The ambivalent relations between RF and BBP are apparent when dealing with the suppression of HWIs. On one hand, improvement of the RF component might save BBP resources. On the other hand, cost-effective BBP solutions for some of the HWIs enable simplification of the RF subsystem. This article presents a system discussion of technical issues for EGPRS PHY design and the trade-offs between RF and BBP implementations. To simplify the discussion, each impairment was dealt with separately. However, these impairments are in many ways interrelated, and they must be jointly compensated to maximize the throughput of an EGPRS DCR-based handset.

About the author
Yuval Dorfan is a DSP engineer for Comsys Communication and Signal Processing. Comsys develops cellular modems for the third generation (EGPRS and UMTS). He received his B.S. from the University of Beer-Sheva in 1998, and his M.S. from the Technion in Haifa in 2000. Dorfan can be contacted at:
yuval.dorfan@comsysmobile.com

References
Namgoong W. and T. H. Meng, ?Direct-conversion RF receiver design,? IEEE trans. On Communications, vol. 49, no. 3, pp. 518-529, March 2001.

Abidi A. A., ?Direct-conversion radio transceivers for digital communications,? IEEE Journal of Solid-State Circuits, vol. 30, no. 12, pp. 1399-1410, December 1995.

Zhang Z., Z. Chen and J. Lau, ?A 900MHz CMOS balanced harmonic mixer for direct conversion receivers, IEEE RAWCON 2000, pp.219-222, Denver, September 2000.

Larson L. E., RF and microwave circuit design for wireless communications, Artech House, Boston, London, 1997.

Crols J. and M. S. J. Steyaert, ?Low-IF topologies for high-performance analog front ends of fully integrated receivers,? IEEE trans. Circuits and Systems - II: Analog and Digital Signal Processing, vol. 45, no. 3, pp. 269-282, 1998.

Banu M., H. Wang, M. Seidel, M. Tarsia, W. Fischer, J. Glas, A. Dec and V. Boccuzzi, ?A BiCMOS double-low-IF receiver for GSM,? Proc. of the IEEE Custom Integrated Circuits Conf., pp. 521-524, May 1997.

Lindoff B., Using a direct conversion receiver in EDGE terminals - A new DC offset compensation algorithm,? PIMRC Conf. September 2000.

Lindoff B. and P. Malm, ?A design rule for DC compensation in direct conversion receivers,? published in the 2001 International Conference on Third Generation Wireless and Beyond.

ETSI, GSM 0505 (release 99), http://www.etsi.org.

Cavers J. K. and M. W. Liao, ?Adaptive compensation for imbalance and offset losses in direct conversion transceivers,? IEEE Trans. On Vehicular Technology, vol. 42, no. 4, pp. 581-588 November 1993.

Proakis J. G., ?Digital communications, third edition,? Mc-Graw-Hill, Singapore, 1995.

ֲ© 2002, PRIMEDIA Business Magazines & Media Inc. All rights reserved. This article is protected by United States copyright and other intellectual property laws and may not be reproduced, rewritten, distributed, redisseminated, transmitted, displayed, published or broadcast, directly or indirectly, in any medium without the prior written permission of PRIMEDIA Business Magazines & Media Inc.