Emerging Technology Series #1

EDGE in Wireless Data

While agreement on a single, worldwide 3G standard is still years away, EDGE modulation – one of the 2.5G proposals – could see wide deployment as an interim technology.

By Hari Shankar

Although most carriers and service providers have plans to deploy enhanced mobile wireless services at speeds that could rival today’s wired telecom data rates, the rollout of high-bandwidth wireless transport technology still faces many uncertainties. On the upside, it is widely agreed that widespread demand will be more than sufficient to support cellular enhancements like high-speed data services and expanded voice capacity. Competitive pressures will also compel service providers to upgrade. Meanwhile, with the standard for a single, universal 3G wireless data transport still years away, the telecommunication industry is growing increasingly impatient to test the world markets for high-bandwidth wireless communication services. Long before any genuine 3G systems come into being, interim 2.5G systems will be rolled out in the coming years based on various existing standards.

he ITU’s IMT-2000 initiative may one day converge, but today three major 3G proposals are still under consideration: cdma2000 (an upgrade to cdmaOne), universal mobile telecom system (UMTS), a wideband-CDMA (W-CDMA)-based approach, and universal wireless communications (UWC)-136. UWC-136 is based on time-division multiple-access (TDMA) as are Europe’s global system for mobile communications (GSM), Japan’s personal digital cellular (PDC), and the digital advanced mobile phone system (D-AMPS) used in the US. Picking the final victor is going to be tough, and a clear-cut winner could take years to emerge. However, if deployment statistics have any bearing, TDMA’s much larger installed base will dictate much of a carrier’s upgrade activity long before we get to 3G.

Existing 2G service providers have already applied to operate 3G networks around the globe. While it’s unclear who will get licenses and what 3G technologies will be adopted, the most likely migration paths are shown in Figure 1 . In the interim, the only 2.5G upgrade close to going live is General Packet Radio Service (GPRS), a packet-switched upgrade for TDMA-based wireless networks (like GSM). In addition, high-speed, circuit-switched data (HSCSD) is another potential upgrade being considered by some GSM networks. Beyond that, enhanced data rates for GSM evolution (EDGE) modulation extensions are planned, which will allow service providers to offer even higher performance, enabling true 3G-like services.

Many potential migration paths exist to reach 3G nirvana. The ITU currently embraces three proposed schemes to attaining the IMT-2000 3G vision, including: UWC-136, an advancement of the US TDMA (IS136) standard that utilizes EDGE technology; W-CDMA (3GPP), a standard from European Telecommunications Standards Institute (ETSI), ARIB, and others; and cdma2000 (3GPP2), the progeny of the cdmaOne (IS95) standard. From TDMA-based 2G providers of GSM and North American dual-mode cellular (NADC) services, interim upgrades will come in the form of GPRS, HSCSD, and IS-136+, and will eventually converge at EDGE for the next throughput upgrade (to 384 kbps) before 3G.

What is EDGE?

EDGE is a new modulation scheme that is more bandwidth efficient than the Gaussian prefiltered minimum shift keying (GMSK) modulation scheme used in the GSM standard. It provides a promising migration strategy for HSCSD and GPRS. The technology defines a new physical layer: 8-phase shift keying (8-PSK) modulation, instead of GMSK. 8-PSK enables each pulse to carry 3 bits of information versus GMSK’s 1-bit-per-pulse rate. Thus, EDGE has the potential to increase the data rate of existing GSM systems by a factor of three.

EDGE retains other existing GSM parameters including a 4.615-ms frame length, eight timeslots per frame, and a 270.833-kHz symbol rate. GSM’s 200-kHz channel spacing is also maintained in EDGE, allowing the use of existing spectrum bands. This fact is likely to encourage deployment of EDGE technology on a global scale.

EDGE system parameters

A traditional 8-PSK system uses raised-cosine filtering to remove intersymbol interference (ISI). With this approach, a one-to- one correlation exists between the required bandwidth and the target symbol rate with no intersymbol interference. For example, a minimum of 270.833 kHz RF bandwidth is required to deploy a system that uses GSM’s 270.833-kHz symbol rate. Although ISI is eliminated using raised cosine filtering, 200 kHz of channel spacing results in a symbol rate less than 200 kHz — the 8-PSK signal with raised-cosine filtering does not fit within 200 kHz of bandwidth.

In order to achieve the desired symbol rate, a more severe filtering approach is required — one that yields poorer performance with regard to ISI. EDGE makes this trade-off, resulting in a conservation of bandwidth at the expense of ISI. This also necessitates a more complex design for the receiver.

Figure 2 is a block diagram of an EDGE modulated system. The serial bit stream is converted into 3-bit words and mapped to the 8-PSK constellation using Gray encoding. The symbols are then rotated by 3p/8 radians to ensure that the envelope of the signal does not go to zero. Next, the symbols are upsampled and filtered using a linearized Gaussian filter (similar to, but different than the method used for GSM). In this manner, the spectrum of the 8-PSK signal can be restricted to 200 kHz.

As seen in the graphs in Figure 3 , linearized Gaussian filtering allows the 8-PSK signal spectrum to occupy the same bandwidth as a GMSK signal. It also introduces a considerable ISI component.

Performance results for an 8-PSK system utilizing linearized Gaussian filtering are illustrated in Figure 3 . In the top graph the 8-PSK signal spectrum is superimposed over the GSM signal (blue). The spectrum of the 8-PSK signal occupies the same bandwidth as a GMSK signal. The bottom plot reveals the degradation in performance with regard to ISI.

By mapping bits to symbols using Gray encoding, adjacent symbols differ by only 1 bit. This minimizes the number of interpreted error bits when a symbol is incorrectly decoded as one of its nearest neighbors.

The envelope of an 8-PSK signal can instantaneously become zero, which complicates the design of the power amplifier. If we continuously rotate the 8-PSK signal constellation by 3p/8 radians, the envelope never goes to zero. This is verified by the circular “clear” area that shows up in the center of the I&Q plot for a signal rotated as described. The rotated constellation is that of a 16-PSK signal. Furthermore, by filtering a 3p/8 radians rotated 8-PSK signal with a linearized Gaussian filter, the signal no longer exhibits a nonconstant envelope — in direct contrast to GSM, which has a constant envelope. The constant envelope, 3p/8 radians rotated 8-PSK signal with a linearized Gaussian filter is shown in Figure 4 . The peak-to-average ratio (PAR) of the signal is 3.3 dB. This makes the design constraints for a power amplifier intended for an EDGE-based system quite a bit more critical than one intended for GSM.

In order to judge the performance of EDGE modulation against GSM, we can compare the bit error rate (BER) of the 8-PSK signal with the BER performance of an MSK signal. This analysis reveals that for low values of Eb/No (less than 10 dB), 8-PSK and MSK provide virtually the same performance. For larger values of Eb/No (above 10 dB), MSK is better than 8-PSK by about 0.25 dB. When the symbol error rate (SER) of 8-PSK is compared with the BER of MSK, it is revealed that a 1-dB penalty is paid for 8-PSK.

The idea behind using the linearized Gaussian filter approach comes from a paper authored by P.A. Laurent, which focuses on representing digital phase modulation by superposition of amplitude modulated pulses. ¹ One fundamental concept in Laurent’s work states that any constant-amplitude binary phase modulation can be expressed as the sum of a finite number of time-limited amplitude modulated pulses. This means that a phase modulator can be approximated using a set of amplitude modulators, where the phase-shaping filter of the phase modulator is replaced by an equivalent set of filters, C _i (t) , for the amplitude modulators. Further, for a large number of phase modulated systems (including GMSK) C ₀ (t) is the most sig-nificant pulse, containing a weighty portion of the energy (for GMSK, over 99% of the energy is contained in C ₀ (t) ). Hence, GMSK can be well approximated by only using the first term of the Laurent expansion.

Letting s(t) represent the complex envelope of a phase-modulated signal with data bits { a _n } and phase pulse w (t) :

(Equation 1)

(Equation 2)

(where the pulses { C _k (t) } are a function of w (t) and the alphabet values { A _k,n } are a function of the data bits { a _n }).

In order to apply Laurent’s results, the frequency pulse of the phase modulator must be finite in duration. An ideal frequency pulse for GSM is non-causal and infinite in duration:

(Equation 3)

where:

(Equation 4)

So, the Gaussian pulse is of infinite duration and must be truncated. For EDGE, the ideal frequency pulse in Equation 3 is truncated from ( -2T, 2 T ), where L = 4 , and made causal. Hence the phase pulse used to derive the EDGE filter is:

(Equation 5)

The impulse response of the linearized Gaussian filter is derived based on this approximation. This impulse response can be expressed as the product of four time-shifted versions of a basic pulse, S(t) . The pulse S(t) is a function of the phase pulse through the phase-modulated system. Based on this phase pulse, the EDGE filter is given by the following equations:

(Equation 6)

where:

(Equation 7)

The EDGE filter equations are programmed using Application Extension Language (AEL), a language similar to Matlab. The generation process begins by calculating the phase function w (t) , from which S(t) is calculated using Equations 6 and 7, and then C ₀ (t) is calculated from S(t) .

A fixed point implementation of the EDGE modulator is shown in Figure 5 . In this design the in-put serial bit stream is converted into 3-bit packets, which are then mapped to the rotated symbols. The mapping pattern between bits and the rotated symbols are stored in ROM, avoiding the need for generating complex exponentials or performing complex multiplications. Filter coefficients are quantized to 8 bits.

If the output spectrum of a floating-point design of the modulator is compared with the fixed-point model, the output of the fixed-point design will contain more noise than the floating-point design.

Because of the severe ISI introduced by the linearized Gaussian filters, an EDGE receiver must include some form of equalizer. For illustration, the very simple receiver in Figure 6 can be analyzed. This system diagram consists of an EDGE modulator, an additive white Gaussian noise channel (AWGN), and a receiver with receive filters and a demodulator (which recovers the bits from the EDGE constellation). The receive filter is designed to remove the ISI introduced by the transmit filter. This kind of filter is known as a zero-forcing filter, hence the receiver is called a zero-forcing receiver. Precise carrier and timing recovery are assumed, and the channel is presumed to have an ideal frequency response.

The receive filter is developed to compensate for the transmitting filter. Further, it is assumed that the frequency response of the AWGN channel is a constant. The demodulation filter is used to make the total frequency response from the transmit-receive filter cascade a constant as well, so that it will yield either no or a controlled amount of ISI.

Hence, if G _T (f) and G _R (f) are the transmit and receive filters and T _symbol is the symbol period, the receive filter is defined by the following equation:

(Equation 8)

The EDGE demodulator recovers the bits from the EDGE constellation by performing the inverse of operations performed by the transmitter. The received symbols are first derotated to obtain 8-PSK symbols. The phases of the symbols are then computed, and a hard decision is made to recover the symbol. The symbols are finally converted into bits using inverse Gray mapping.

Analysis of BER performance of the zero-forcing receiver reveals that there is a loss of 7 dB in performance compared to ideal 8-PSK, meaning that better receivers are required. Other equalization schemes such as a decision feedback equalizer and a Viterbi equalizer are under current investigation.

Migration strategy

The EDGE modulation scheme represents a possible migration strategy for 2.5G HSCSD and GPRS services. EDGE has the potential to increase the data rate of these GSM-based services by a factor of three, attaining near 3G performance levels.

Hari Shankar is an applications engineer developing applications focused on communication systems and digital signal processing. He received his BSEEfrom IITKanpur and his PhD from the University of Hawaii. He can be reached at hshankar@dspc.com.

Return to the Table of Contents