DSP-BASED FM STEREO GENERATOR FOR DIGITAL STUDIO -TO - TRANSMITTER LINK

Transcription

1 DSP-BASED FM STEREO GENERATOR FOR DIGITAL STUDIO -TO - TRANSMITTER LINK Michael Antill and Eric Benjamin Dolby Laboratories Inc. San Francisco, Califomia ABSTRACT The design of a DSP-based composite stereo generator for use in a digital studio-to-transmitter link is described. The functions of pilot and subcarrier generation, mutrixing, modulation, and test tone generation are all accomplished by means of a single DSP chip, resulting in a high performance, low complexity implementation. Performance of the algorithm is nearly ideal, and signal quality is shown to be dependant only on the digital to analog conversion process. 1. Introduction Digital signal processing (DSP) technology has revolutionized all aspects of audio production and reproduction, including the commercial broadcast environment. Many signal processing tasks previously accomplished with analog techniques can now be implemented in the digital domain with reduced complexity and higher performance. The studio-totransmitter link (STL) is one example of this transition; although traditionally an analog function, the selective application of DSP can make simultaneous improvements in audio quality, performance stability, and RF spectral efficiency. Digital audio coding and modulation techniques have been employed in the design of a new 950MHz STL for FM broadcast, incorporating a DSP-based FM stereo generator with its attendant advantages. Figure 1 is a receiver block diagram for the digital STL (DSTL ), showing the individual receiver/ synthesizer, demodulator, audio decoder, D/A converter, alarm, power supply, and stem generator modules. The latter uses a single programmable digital signal processor to generate a composite stereo signal from the PCM audio output of the audio decoder module; as a result, the audio signal remains in the digital domain until final conversion to composite analog. The benefits of this approach include superior audio quality, improved stability of operation, and the higher reliability and lower costs associated with a low complexity design. 2. Functional Description The basic function of an FM stem generator is to provide a composite output signal consisting of the baseband (sum of left and right input channels), a 19 khz pilot signal, and a 38 khz suppressed-carrier subchannel, amplitude modulated with the difference of the input channels (L-R) [2]. FREQUENCY (Hz) Figure 2 Figure 2 depicts the spectrum of the composite stereo signal based on a uniform spectral distribution in the input signal $ IEEE 622

2 Previous implementations of this function have centered on two basic analog techniques [3,4,51. In matrix-type generators, baseband and &ffe&nce &nals are generated by precision analog networks. The difference (L-R) signal and 38 khz subcarrier are then fed to a balanced modulator (typically an analog multiplier) to create the double-sideband suppressedcarrier subchannel centered at 38kHz. Baseband, subchannel, and pilot signals are then summed to produce the composite output. In switching-type generators, an electronic switch is used to alternately select the left and right input channels at a 38 khz rate, followed by low-pass filtering to remove harmonics of the subcarrier. Other circuit topologies have also been implemented which combine aspects of both matrix and switching generators in order to exploit the respective advantages of each technique. In all these approaches, performance is limited by a number of sources including component tolerances, nonlinearities, noise, temperature variation, and aging effects. Audio parameters subject to variation andor degradation include distortion and dynamic range, channel separation, and the suppression of spurious modulation products. In order to minimize these effects, DSP techniques have been employed in the design of a new stereo generator. Figure 3 As shown in Figure 3, a single %-bit Motorola DSP56001 makes up the signal processing core of the module, followed by a high performance digital-toanalog converter, output filter, and buffer amplifier. System alignment and test is simplified by a user interface that directly controls the operation of the processor. The composite signal is generated entirely in the digital domain before conversion to analog output. 3. Description of Operation Figure 4 provides a flow diagram of DSP operations in the digital stereo generator. 16-bit digital audio data at a sample rate of 44.1 khz is received from the DSTL audio decoder into the serial input port of the DSP The processor then calculates sum (L+R) and difference (L-R) signals from the left and right channel inputs, with the ability to scale the difference signal under user control. This in turn allows the L-R component of the composite output to be adjusted to an accuracy of f.005 db, in order to compensate for amplitude errors in the FM exciter response. Figure 4 Before modulation of the 38 khz subcarrier, it is first necessary to upsample the sum and difference signals to accommodate the subcarrier bandwidth and avoid alias products in the modulated output. Since the modulation process results in a 53lcHz signal bandwidth, it is sufficient to upsample by a factor of 4 to IrHz. The oversampling process is the most computationally intensive part of the DSP algorithm, and removes L+R and L-R signal images below 88.2 khz by means of a two-stage linear phase F'IR interpolation filter. Since the input signals have been previously bandlimited to 15kHz by the audio coding process, sharp transition bands are not required in the oversampling filters. The resulting specifications can be met by cascaded halfband filters [6], which serve to minimize the processor workload. These structures are characterized by alternating zero-valued coefficients, which reduce the total multiply-accumulate operations by nearly one-half. Filter response is characterized by passband ripple less than.001 db and image attenuation greater than 90 db. A second-order IIR oscillator is used to generate the 19kHz pilot reference signal. The output of the 623

3 oscillator is also squared, offset by one-half, and scaled by two to provide a phase-locked 38 lchz subcarrier. In order to produce the correct phase relationship between pilot and subcarrier signals, pilot phase is shifted by linearly combining the present and previous output samples of the oscillator. The subcarrier is multiplied by the scaled difference signal to produce the 38 khz double-sideband subchannel, which is then added to the baseband &+R) and user-scalable pilot signals. The resulting composite signal is passed to the D/A converter via the output serial port of the processor. All of the tasks of pilot and subcarrier generation, modulation, composite summation and output take place at the upsampled rate of khz. For system set-up, alignment, and test purposes, an audio test oscillator is implemented in the same manner as the pilot oscillator. Under user control, reference level tones at 400 Hz and 1 Wz can be substituted for both input channels, the right channel input can be disabled or inverted, and the pilot signal can be independently turned off. These controls facilitate adjustment of operating levels, optimization of channel separation, and the measurement of various performance parameters. For flexibility of operation, the generator also allows remote or local selection of stereo (composite) or mono (baseband only) output. Performance of the DSP algorithm is nearly ideal, and the composite PCM signal is limited almost entirely by the accuracy of the input signals. Figure 5 displays the spectrum of the 16-bit composite output, using the intemal 1 Wz test oscillator at a -2 dbfs level into the left channel only. In this case, input resolution is not limited to 16 bits, and the error spectra are due primarily to residual images of the oversampled input signal. The matrixing and modulation operations are linear to within the 24-bit arithmetic precision of the processor, and oscillator stability is crystal-based. Consequently, it is imperative that the D/A converter and output filter approach this level of precision as closely as possible in order to minimize any loss of performance in the conversion to composite analog. Conversion to a quantized analog signal is performed by a 16-bit R-2R D/A converter designed for oversampling use. This part is characterized for high frequency operation and exhibits fast settling at the composite sample rate of 176.4kHz. Although performance is excellent in oversampled applications, distortion increases rapidly above 20Wz due to intermodulation (IM) in the internal current to voltage converter. It is therefore necessary to bypass this circuit with an external high-speed amplifier having a faster slew rate with reduced IM. By integrating the signal slightly, this stage also reduces the slew rate requirements for subsequent circuitry. Following digital-to-analog conversion, the composite signal is filtered in order to remove images occurring above Wz (176.4 khz - 53 khz). This filter must meet precise magnitude and phase specifications in order to maintain high separation between the multiplexed channels [3]. In particular, the overall frequency response must be flat to within k 0.17 db of magnitude and k 0.11 degree of linear phase in order to achieve a channel separation of 60 db throughout the audio band. The signal is initially filtered by a ninth order elliptic reconstruction filter with.01 db passband ripple and an 85 Wz band edge. Attenuation of images is greater than 84 db for frequencies above 120 WZ. This filter is followed by a second order high frequency boost to pnxisely compensate for the sinx/x loss due to the finite sample length of the D/A converter (1.43 db at 53 IrHz). The phase linearity of these networks is excellent within the composite signal passband (0-53 khz), and the overall phase response is held to k 0.1 degree with the aid of a second order phase compensator. This all-pass network supplies an additional 360 degrees of total phase shift, with a slope that compensates for the phase error accumulated in the elliptic filter and sinx/x compensator. After filtering, the composite output is buffered by a two-stage amplifier with a frequency response of k.002 db from 0 to 53 W z and twin tone IM distortion below -100 db. The design is capable of driving loads as low as 50 ohms, and DC offset at the output is held to within 2 mv. The buffered composite signal and its return are coupled to the output connector via a two pole form C relay that engages when the DSTL receiver is in the operate mode. 4. Performance The performance of the DSP algorithm is illustrated in Figures 5, 6, and 7, which display the spectrum of the composite PCM output for various combinations of left and right channel inputs using the 1kHz digital test oscillator. Maximum input level in the digital domain is -2dBFS to avoid the possibility of clipping. All of the spectra are limited to 88.2Wz, the Nyquist rate of the oversampled output. 624

4 ; LEVEL -60 (a) 80 -im 0 20k 4Ok 60k 8Ok 1Wk Figure 5 Figure 5 represents the case of a single channel input signal (left only), which produces equal energy in both the baseband and modulated subcarrier signals. The primary artifacts are images of the baseband component at 87.2kHz and of the subcarrier modulation at 49.2 khz and khz. Image attenuation is a function of the oversampling filter reject-band performance and exceeds 100 db. LEVEL : (a) 80! I I 0 20k 40k 60k 8Ok look Figure 6 Figure 6 displays the condition of identical left and right channel inputs (mono operation), producing a full level baseband signal. The only significant artifact is the baseband image at 87.2 khz at a level of -105 dbfs LEVEL bo (a) k 40k 60k 8Ok 100k Figure 7 Figure 7 shows the case of equal level but inverted polarity left and right channel inputs, which produce a full level modulated subcarrier. Images of the sidebands are present at 49.2 khz and lchz and are attenuated by more than 100 db. Figures 8, 9, and 10 represent generator performance at the composite analog output for the same input conditions as figures 5, 6, and 7, respectively. These spectra have been rescaled such that full scale represents 100% modulation. Performance is limited by the D/A converter and output filter circuits, where degradation is primarily caused by intermodulation and harmonic distortion. The composite stereo signal is particularly sensitive to nonlinear distortion due to its multiplexed nature, and distortion can result in nonlinear crosstalk in the demodulated output. Figure 8 Figure 8 shows the analog spectrum for the case of a single channel input. Distortion artifacts include third order sidebands of both pilot and subcarrier modulation, and second and fourth harmonics of the pilot. Signal levels are below -90 db for each of these components LEVEL (a) r, I 0 20k 40k 60k 8Ok look Figure 9 Figure 9 represents the mono WR) operating condition. The second harmonic of the test tone represents the only significant distortion product at a level of - 92 dbfs k 40k 60k 8Ok 1Wk FREQUENCY e) Figure 10 Figure 10 displays difference signal (L-R) modulation. Distortion products include sidebands of the subcarrier modulation, and the fourth harmonic of the pilot at 76 khz. Each of these is more than - 90 dbfs. 625

5 Care must be taken to differentiate the performance of the generator from that of the measurement system - in this case, an FFT spectrum analyzer with a 16-bit A D converter. For this reason, the spectra presented here are concatenations of separate baseband and subchannel measurements in which high quality analog filters were used to suppress high level signal components before analysis. The spectral noise floor displayed in the figures is also limited by the analyzer, except in the region of 80 khz where additional noise is added by the generator output filter. The performance evaluation of any stereo generator is complicated by the fact that many specifications can only be measured accurately after decoding, and will therefore incorporate the performance of the decoder. Noise and distortion can be evaluated directly from the output spectrum, although the quality of the analyzer will be a limiting factor. By comparison, separation is a sensitive parameter that depends not only on the magnitude and phase response of the generator, but on the corresponding response of the FM exciter and its connecting cable. In actual practice, separation is often measured and optimized by means of the stem monitor used in a particular installation. Loss of separation can also occur during transmission (via multipath distortion), and during reception and demodulation. Since degradation can occur at each step in the chain, high performance in the generator is particularly impomt. 5. Conclusion 6. Acknowledgement The authors wish to acknowledge the conmibutions of Mark Atherton and Grant Davidson to the definition of the digital stereo generator, and to its algorithm, software, and hardware development. 7. References 1. R. Bell, "RF Design Considerations in the Development of a High-Spectral Efficient, Multi-Channel, All-Digital STL," Proc., 46th Annual NAB Broadcast Eng. Conf., Las Vegas, A. Csicsatka and R. Lintz, "The New Stereo FM Broadcasting System- How to Understand the FCC Specifications and Generate the Composite Signal," J. Audio Eng. Soc., vol. 10, pp. 2-7, Jan T. Brook, "Stereo Coder, Part 1," Wireless World, vol. 83, pp , Apr D. Hershberger and G. Mendenhall, "New Techniques for Generation of Composite Stereo Signals," Proc., 31st Annual NAB Broadcast Eng. Conf., Washington, A.J. Oliveira, "A Design Method for High-Audio-Quality FM Multiplex Encoders," J. Audio Eng. Soc., vol. 40, pp Mar F. Mintzer, "On Half-Band, Third-Band, and Nth-Band FIR Filters and Their Design," EEE Trans. on Acous., Speech, and Sig. Proc., vol. ASSP-30, pp , Oct DSP technology has made possible the design of a new digital studio-to-transmitter link for radio broadcast, including an FM stereo generator in which the composite signal is synthesized entirely in the digital domain. In contrast to traditional approaches, critical signal processing tasks are performed by a single DSP chip. The result is a high performance, low complexity implementation that minimizes the limitations of analog circuitry. Algorithm performance is nearly ideal, and signal quality is dependant only on the digital to analog conversion process. Typical distortion products using a 16-bit D/A converter are more than 90 dl3 below maximum modulation, resulting in excellent system performance. 626