`
`Amit Bodas
`Digital Wireless Communications Laboratory
`Dept. of Electrical and Computer Engineering
`University of California, Davis, CA 95616
`Tel: (916) 752-4608; FAX: (916) 752-8428
`Email: bodas @ece.ucdavis.edu
`
`Kamilo Feher, Ph.D., Fellow, IEEE
`ECE. Dept. UC. Davis, Davis CA 95616
`and Vice President, Consulting and Licensing Group,
`Digcom, Inc.
`44685 Country Club Dr.
`El Macero, CA 95618.
`Tel: (916) 753-0738; FAX: (916) 753-1788
`
`Abstract --- Reduced complexity Integrated Circuit (IC) design
`architecture for constant envelope modulation is presented.
`This method is demonstrated on the globally standardized
`Gaussian Minimum Shift Keying (GMSK) for the “Global Sys-
`tem for Mobile Communications” (GSM). The in-phase and
`quadrature signals are generated directly from the input bi-
`nary data, bypassing the separate steps of integrating, and then
`calculating the sine and cosine values. The implementation
`takes advantage of waveform symmetry properties thereby
`leading to reduction in memory requirements by 800% with
`respect to full storage of required waveforms. Performance
`comparison with a commercially available GSM IC is also car-
`ried out.
`
`I. INTRODUCTION
`
`The Gaussian Minimum Shift Keying (GMSK) modula-
`tion[1,2] scheme is very suitable for mobile radio applica-
`tions due to the following features,
`(i) Compact output spectrum which results in increase in
`number of available channels and at the same, maintain low
`adjacent channel interference.
`(ii) Constant Envelope which allows RF amplifiers to operate
`in non linear mode with greatly improved power efficiency.
`This is very important in today’s wireless cellular PCS sys-
`tems for battery powered units.
`In the following sections, conventional GMSK technique
`is briefly introduced and an alternative method for generat-
`ing GMSK signals is described. The results of the measure-
`ments are then discussed.
`
`A. Gaussian Minimum Shift Keying Modulation
`
`Figure 1 shows the textbook architecture for the quadra-
`ture baseband processor. Input NRZ data is filtered using a
`Gaussian low pass filter (GLPF). This filter may be imple-
`mented using an actual filter that approximates the Gaus-
`sian characteristics or using the ROM look up table method.
`
`This signal after integration ( b(t) ) is split and fed into co-
`sine and sine look up tables such that the input signal drives
`are related by the equations,
`
`x(t) = cos[ b( t) ]
`y(t) = sin[ b( t) ]
`
`(1)
`(2)
`
`Evidently the I and Q signals are mathematically related
`or “cross correlated” [3,4] to each other. Thus in this case
`the predictability or cross correlation of y(t) from x(t) is de-
`fined by,
`
`This is also evident from figure 2 which shows the wave-
`forms for the I and Q channels. It can be observed that when
`the quadrature (or in-phase) signal reaches its maximum
`magnitude, the in-phase (or quadrature) signal is crossing
`zero. The amplitude of the quadrature (or in-phase) signal is
`reduced when the in-phase (or quadrature) signal is non
`zero. This cross correlation property maintains the constant
`envelope.
`
`4
`
`sin [ b(t) ]
`
`Integrator
`
`y(t) = sin[b(t)l
`
`Figure 1. Textbook Implementation of GMSK Baseband Processor
`
`0-7803-3302-0196 $5.00 0 1996 IEEE
`
`103
`
`Qualcomm Incorporated
`Exhibit 1015
`Page 1 of 4
`
`
`
`I
`
`Max
`
`Max
`
`Data
`
`[03 02 D1 DO]
`
`Flag
`Generator
`
`Wavetwm
`Select
`Logic
`
`2
`
`~~
`
`Add(
`Gen
`
`ROM
`
`-
`
`+
`+
`
`I
`
`Figure 3. Block diagram of the GMSK modulator
`
`A ) ROM Look Up table and Waveform Select Logic
`
`Advantage of the symmetry of the waveform is taken.
`Thus only 4 waveforms as shown in Figure 4 need to be
`stored in the ROM Look Up Table [6]. Each waveform con-
`tains 8 samples per bit duration. By inverting the stored
`waveform, memory reduction by a factor of 2 is achieved.
`Time reflecting the waveform reduces the memory require-
`ment by another factor of 2. Also since 16 samples are used
`per symbol, and we use 8 samples per bit duration, a further
`reduction by a factor of 2 is achieved. The stored waveforms
`are 10 bits wide and 8 times oversampled (per bit duration)
`
`TABLE I
`
`Wavelets to be selected for I and Q channels based on F D3 0 2 D1 DO (Read
`in this order under each set). The incoming bit is D3 while DO is the input data
`4 bit times prior. In each set the top wavelet is for I channel while the wavelet
`below for Q channel [5]
`
`rprprtd
`
`00100
`
`00101
`
`00110
`
`00111
`
`Figure 2. Cross correlated I and Q channel signals.
`
`11. MODULATOR IMPLEMENTATION
`
`Due to the cross correlation properties of the I and Q
`signals, it is possible to pre-calculate the output waveform in
`response to the input data. Four serial data inputs D3
`through DO (D3 is the incoming bit while DO is the input
`data 4 bit duration's prior) and one Flag bit 'F' are used for
`the waveform selection procedure. The Flag bit alternates
`between a 1 or 0 per bit period and provides information as
`to which half of the symbol waveform to select. When F = 0,
`the first half of the symbol waveform is selected for the I
`channel while for the Q channel the second half is selected.
`In addition when F = 0, DO and D2 provide data transition
`information for the I channel while D1 and D3 provide data
`transition information for the Q channel. For the I channel,
`when F = 0, D2 is the current bit and DO is the past bit. For
`the Q channel, D1 is the current bit and D3 is the next in-
`coming bit. If the current bit is 1, then the respective wave-
`form is of positive amplitude else of negative amplitude. If
`the I channel (DO to D2) and Q channel (D1 to D3) show no
`data transition then the respective waveforms begin and end
`at intermediate values else they will begin or end at k 1 or 0.
`
`With F = 1, the relationship of DO through D3 with the I
`and Q channels is reversed. Now D1 and D3 provide infor-
`mation for the I channel while DO and D2 provide informa-
`tion for the Q channel. The 32 different wavelet pairs in re-
`sponse to 4 serial data inputs and Flag are shown in Table H
`[5]. For each set in Table I, the top wavelet is for the 1 chan-
`nel and the one below it is for Q channel.
`
`Thus using Table I, the output waveforms for the I and Q
`channel are selected in response to 4 data input bits and one
`Flag bit. Thus the GMSK modulator can be implemented as
`a ROM Look Up table system as discussed further. The block
`diagram of our implementation is shown in Figure 3 fol-
`lowed by the baseband processor description.
`
`104
`
`Page 2 of 4
`
`
`
`which requires a ROM of 4 waveforms x 10 bits/ sample x 8
`samples = 320 bits per channel. With no waveform reduc-
`tion, 16 symbol duration waveforms will be required. Hence
`a memory reduction of 8 times (800%) with respect to the
`full symbol storage is gained.
`
`In response to the 4 input data bits and 1 Flag, the I and Q
`waveform select logic selects one of the 4 waveforms and
`generates control signals to invert and or time reflect the
`selected waveform. Time reflection is achieved by the count
`DOWN mode of the address generator.
`
`1
`
`0 9
`
`0 8
`
`0 7
`
`0 6
`
`9 H0
`
`0 4
`
`03
`
`0 2
`
`01
`
`' 0
`
`005
`
`0 1
`
`015
`
`025
`0 2
`Bit Ouralwn
`
`0 3 035 0 4
`
`045
`
`0 5
`
`Figure 4. Waveforms stored in the ROM for BTb = 0.3 with A = 0.52 and
`B = 0.85. 8 samples are used per bit duration waveform.
`
`B) Data Encoder
`
`To ensure that the output data pattern is identical to that
`generated by a differentially encoded GSM modulator, a data
`encoder is designed. Depending on present state (Sn.l) and
`incoming bit (Dn), the next state (S,) will be determined as
`per the state diagram shown in Figure 5. The 1 or 0 on the X
`or Y axis denotes the output of the state. If the present state
`is initially on the X axis (Figure 5a) then during the next bit
`time, the present state will be on the Y axis (Figure 5b). The
`Flag bit which alternates per bit time is used to control the
`output of the data encoder as shown in Figure 6.
`
`Data In (Dn)
`(270.833 Kb/$) '
`
`Encoded Data Output (Sn)
`
`Figure 6. Data Encoder Logic Circuit.
`
`111. DISCUSSION OF RESULTS
`
`For measurement purposes, a baseband processing system
`is designed and implemented with a Field Programmable
`Gate Array (PGA). The design utilizes about 99 out of a
`maximum 196 Configurable Logic Blocks (CLB) which is
`about 50% device utilization. The GSM data rate of 270.833
`Kb/s generated by a pseudorandom data generator is used.
`Results are compared to those generated with a commercially
`available GSM IC.
`Figure 7 shows the output waveforms generated by the
`designed system (Figure 7a) and the GSM IC (Figure 7b) for
`the same input data pattern.
`
`Figure 7a
`
`Figure 7b
`
`Figure 5a. State Diagram
`beginning on X axis .
`
`Figure 5b. State Diagram
`beginning on Y axis.
`
`Figure 7. Output I and Q baseband waveforms of the designed baseband proces-
`sor (a ) and of a commercially available GSM IC (b ). The same input data
`pattern (shown in the middle) is used.
`
`105
`
`Page 3 of 4
`
`
`
`Figure 8 shows the constellation of the I and Q signals.
`The output signals have the same amplitudes such that the
`vector sum of the I and Q signals are approximately the same
`at virtually all phases of each bit period.
`
`10”
`
`\\
`
`I
`
`6
`
`7
`
`8
`
`9
`
`10
`
`11
`S/N dB
`
`12
`
`13
`
`14
`
`15
`
`16
`
`Figure 10. Pe versus S/N for the GMSK baseband processor (Curve 2) and
`GSM IC (Curve 1). The post-detection filter used is a 4* order Butterworth
`filter with BiT = 0.6.
`
`IV. CONCLUSIONS
`
`In this paper, a design and implementation of a hardware
`efficient quadrature GMSK structure based on the cross-
`correlation concepts is discussed. Waveform symmetry leads
`to reduction in memory requirements. The encoder allows
`the data polarity to match exactly with the output polarity of
`the commercial GSM IC. Performance results are compared
`with a commercially available GSM IC. The modulator de-
`sign presented here is suitable for IC implementation at vari-
`ous levels.
`
`V. REFERENCES
`
`[ 13 K. Feher, “Wireless Digital Communications: modula-
`tion and spread spectrum applications,” Prentice Hall, 1995.
`[2] K. Murota and K. Hirade, “GMSK modulation for digital
`mobile radio telephony,” IEEE Transactions on Communi-
`cations, Vol. COM-29, pp. 1044-1050, July 1981.
`[3] S. Kato, K. Feher, “ Correlated Signal Processor,” U.S.
`Patent 4.567.602, issued January 28, 1986.
`[4] S. Kat0 and K. Feher, “XPSK: a new cross-correlated
`phase shift keying modulation technique,” IEEE Transac-
`tions on Communications, pp. 701-707, May 1993.
`[5] H. Yan, “Wavelet and Logic for GMSK BT = 0.5 DSP
`Implementations,” Digital & Wireless Communications Lab.
`Report, UC. Davis, April 13, 1996.
`[6] D. Wyskiel, “Baseband Modem Design and Architecture
`to Support Multiple Wireless Applications,” M.S. Thesis,
`University of California ,Davis, 1996.
`[7] GSM 5.05, “European Digital Cellular Telecommunica-
`tion Systems (Phase 2): Radio Transmission and Reception.”
`ETSVGSM. May 1994.
`
`Figure 8. Constellation Diagram of the designed system
`
`Figure 9 shows the superimposed power spectral density
`(PSD) results for the designed GMSK baseband processor
`and the commercial GSM IC. The input bit rate is 270.833
`Kb/s. The GSM spectrum mask [7] is also shown. The meas-
`ured spectrum does not go over the GMSK mask. Thus the
`baseband processor meets the GSM spectrum mask.
`
`Figure 9. Measured Power Spectral Density. The bright line is the commercial
`GSM IC while the light trace is the designed GMSK baseband processor.
`GMSK mask is also shown.
`
`Figure 10 presents the measured probability of error (P,)
`versus the S/N relationship, where S/N is the average signal
`power to average noise power at the decision threshold in-
`put. Measurement for single channel is done where the B,T
`of the post-detection filter is 0.6. The difference between the
`BER performance of the designed baseband processor (Curve
`2) and the commercial GSM IC (Curve I) at Pe =
`is
`measured to be under 0.5 dB.
`
`106
`
`Page 4 of 4
`
`