5
A FREQUENCY-SELECTIVE, DIGITALLY IMPLEMENTED DISTURBED-CHANNEL SIMULATOR Bruce J. Currivan Systems Engineering Manager Stanford Telecom 2421 Mission College Boulevard Santa Clara, CA 95054 ABSTRACT The Dispersive Channel Simulator (DCS) was developed to evaluate the performance of adaptive-equalizer- based receiving equipment and mission data processing software in the presence of frequency-selective or flat ionospheric scintillation. It operates on digital input data to produce scintillated outputs on three indepen- dent channels. The time-varying channel effects consist of linear filtering, total-electron-content (TEC) envelope delay, TEC carrier frequency modula- tion, signal absorption loss, and additive noise includ- ing sky noise temperature. Parameters for a one-hour scenario are read from a cartridge tape. The channel convolvers contain 96 complex taps per diversity channel with a sample rate of 10 Msps. Modulation is BPSK or QPSK with selectable convolutional encoding. The simulator hardware is housed in a 15.75-inch chassis and is controlled by a TEMPEST PCIAT computer. software which processes the receiver output also makes use of the DCS to produce simulated outage conditions. Previous dispersive channel simulators were imple- mented with analog tapped delay lines or surface acoustic wave (SAW) convolvers. The recent advent of high-speed digital finite-impulse-response (FIR) filter integrated circuits has made it practicable to imple- ment a 3-diversity-channel, digital, frequency-selec- tive disturbed channel simulator with 96 complex taps per channel at 10 megasamples per second in a small physical space. CHANNEL MODELS As shown in Figure 1, the DCS is driven by channel realization parameters computed offline and stored on a scenario cartridge tape. The tape contains files of channel impulse response functions (CIRFs), TEC delay, TEC doppler, signal attenuation, and noise attenuation. All parameters are time varying, with sufficient data on the tape for a l-hour simulation. Figure 1. Dispersive channel simulator channel effect data flow diagram INTRODUCTION Dispersive channel simulators (DCS) are needed for testing communications links in the presence of frequency selective fading, in which the channel symbol rate exceeds the channel frequency selective bandwidth, fO. Adaptive-equalizer-based receivers [l], which are meant to operate in severe frequency-selec- tive fading, require a DCS as a test bed. Mission The CIRFs may be computed using the Antenna Channel Impulse Response (ACIRF) program developed by DNAIMRC [Z]. ACIRF files are generated based upon selected values of decorrelation time (tauo), frequency-selective bandwidth (fO), decorrelation distance (LO), and antenna geometry. The resulting ACIRF output files consist of 1024 samples of the CIRFs on each diversity channel, representing 102.4 tau0 of channel time. 29.1.1. CH2831-6/90/0000-0657 $1 .OO 0 1990 IEEE 0657

[IEEE IEEE Conference on Military Communications - Monterey, CA, USA (30 Sept.-3 Oct. 1990)] IEEE Conference on Military Communications - A frequency-selective, digitally implemented

  • Upload
    bj

  • View
    212

  • Download
    0

Embed Size (px)

Citation preview

Page 1: [IEEE IEEE Conference on Military Communications - Monterey, CA, USA (30 Sept.-3 Oct. 1990)] IEEE Conference on Military Communications - A frequency-selective, digitally implemented

A FREQUENCY-SELECTIVE, DIGITALLY IMPLEMENTED

DISTURBED-CHANNEL SIMULATOR

Bruce J. Currivan Systems Engineering Manager

Stanford Telecom 2421 Mission College Boulevard

Santa Clara, CA 95054

ABSTRACT

The Dispersive Channel Simulator (DCS) was developed to evaluate the performance of adaptive-equalizer- based receiving equipment and mission data processing software in the presence of frequency-selective or flat ionospheric scintillation. I t operates on digital input data to produce scintillated outputs on three indepen- dent channels. The time-varying channel effects consist of linear filtering, total-electron-content (TEC) envelope delay, TEC carrier frequency modula- tion, signal absorption loss, and additive noise includ- ing sky noise temperature. Parameters for a one-hour scenario are read from a cartridge tape. The channel convolvers contain 96 complex taps per diversity channel with a sample rate of 1 0 Msps. Modulation is BPSK or QPSK with selectable convolutional encoding. The simulator hardware is housed in a 15.75-inch chassis and is controlled by a TEMPEST PCIAT computer.

software which processes the receiver output also makes use of the DCS t o produce simulated outage conditions.

Previous dispersive channel simulators were imple- mented with analog tapped delay lines or surface acoustic wave (SAW) convolvers. The recent advent of high-speed digital finite-impulse-response (FIR) filter integrated circuits has made i t practicable to imple- ment a 3-diversity-channel, digital, frequency-selec- tive disturbed channel simulator with 96 complex taps per channel at 10 megasamples per second in a small physical space.

CHANNEL MODELS

As shown in Figure 1, the DCS is driven by channel realization parameters computed offline and stored on a scenario cartridge tape. The tape contains files of channel impulse response functions (CIRFs), TEC delay, TEC doppler, signal attenuation, and noise attenuation. All parameters are time varying, with sufficient data on the tape for a l-hour simulation.

Figure 1. Dispersive channel simulator channel effect data flow diagram

INTRODUCTION

Dispersive channel simulators (DCS) are needed for testing communications links in the presence of frequency selective fading, in which the channel symbol rate exceeds the channel frequency selective bandwidth, fO. Adaptive-equalizer-based receivers [l], which are meant to operate in severe frequency-selec- tive fading, require a DCS as a test bed. Mission

The CIRFs may be computed using the Antenna Channel Impulse Response (ACIRF) program developed by DNAIMRC [Z]. ACIRF files are generated based upon selected values of decorrelation time (tauo), frequency-selective bandwidth (fO), decorrelation distance (LO), and antenna geometry. The resulting ACIRF output files consist of 1024 samples of the CIRFs on each diversity channel, representing 102.4 tau0 of channel time.

29.1.1. CH2831-6/90/0000-0657 $1 .OO 0 1990 IEEE 0657

Page 2: [IEEE IEEE Conference on Military Communications - Monterey, CA, USA (30 Sept.-3 Oct. 1990)] IEEE Conference on Military Communications - A frequency-selective, digitally implemented

Channel simulations typically fall into three categories: additive white Gaussian noise (AWGN), flat fading, and frequency-selective fading.

The AWGN channel results when no disturbance is present other than thermal noise. A single tap in each channel of the FIR filter is nonzero and has a constant magnitude, controlled by the signal attenuation parameter, and constant phase.

Flat fading results when a single tap in each channel is active with random magnitude and phase. Typically the I and Q tap weights are independent Gaussian, with average rate of change proportional to l/tauO. The signal phase is then uniformly random and the amplitude follows the Rayleigh distribution, with arbitrarily deep fades occurring, limited by sampling and quantization effects and by the accuracy of the vector modulator. The depth to which accurate fading fidelity must be preserved depends on the link being modeled. Depths of 40 dB are readily attainable.

FREQUENCY-SELECTIVE FADING

Frequency-selective fading corresponds to multiple active taps in the FIR filter. This occurs when fO is less than the channel symbol rate. Each tap is normally modeled as Rayleigh. Any degree of correla- tion may occur between the CIRFs on the three diver- sity channels, as determined by the geometry of the link and the antenna pattern.

Frequency-selective fading is of great interest in receiver characterization. The multiple active taps cause intersymbol interference which is catastrophic for standard coherent receivers. Adaptive-equalizcr- based receivers are able to remove the intersymbol interference and take advantage of the diversity inherent in the channe:. In the time domain, diversity results from the fact that since many decorrelated taps are active in the FIR filter, the probability that all will fade simultaneously is small. In the frequency domain, the random CIRF transforms into a random spectrum containing peaks and nulls. Frequency diver- sity results from the fact that some frequency bands are likely to be non-faded a t any given time. Deep fades are thus less likely in frequency-selective fading than in flat Rayleigh fading.

DIGTTAL DATN - CLOCK IN

DIGITAL DATAICLOCK

OUT

Frequency-selective fading contains signal variations on three disparate time scales. First, signal-to-noise ratio is affected by atmospheric absorption and sky noise, with times on the order of tens of seconds to hours. Second, the signal envelope fades in and out as the CIRF changes, with times characterized by tauO ranging from milliseconds to seconds. Third, inter- symbol interference causes the instantaneous signal amplitude to vary with times on the order of from less than a symbol to a few symbols.

CHANNEL PROCESSING

After generating the CIRF realizations, the raw CIRF files are processed to produce a scenario tape suitable for playback on the DCS. The CIRFs are normalized to a constant average power in order to remove reali- zation-dependent gain variation. The CIRFs are then quantized to 16 bits on each component, real or in- phase (I) and imaginary or quadrature (Q).

Since the computation time and storage requirements for the CIRF files are limiting factors in a long simulation, the DCS uses each CIRF file over and over until the next CIRF file is loaded. A recirculation memory is provided for this purpose as shown in Figure 1. This requires that the CIRF files be circularly continuous, which is the case for ACIRF realizations. A new CIRF file can be loaded into the recirculation memory every 20 seconds, determined by the computer's mass storage transfer speed. To smooth the discontinuity when a new CIRF file is loaded, 32 transition CIRFs are read out from a transition memory before beginning the next CIRF file. The transition CIRFs are computed off-line using gradually decreasing (cosine) weighting of the previous CIRF and gradually increasing (sine) weighting of the new CIRF, in order to keep the expected power constant during the transition.

A new CIRF is transferred from the recirculation or transition memory to each FIR filter 10 times per decorrelation time (tauO). The other scintillation parameters are updated every 5 seconds from the parameter memory. The minimum tau0 is 1 milli- second, limited by the transfer rate into the FIR filters. A t a sample rate of 10 MHz, the minimum f O that can be simulated in the DCS is 65 kHz, deter- mined by the number of taps (96).

ANALOG 1 L Q BASEBAND (€4 M M

OUT IF OUT 4 4 4 A A A

S-BAND OUT 2.2 GHz

I I OMA INTERFACE

SCINTILLATION GENERATOR UNIT t t t

SCINTILLATION

ACIRF CHANNELS, CARTRIDGE TEMPEST

DOPPLER, TAPE C P U DELAY, 4-b SERIAL CONTROUSTATUS

CINO ~ - Figure 2. Dispersive Channel Simulator Functional Block Diagram

29.1.2. 0658

Page 3: [IEEE IEEE Conference on Military Communications - Monterey, CA, USA (30 Sept.-3 Oct. 1990)] IEEE Conference on Military Communications - A frequency-selective, digitally implemented

SIMULATOR ARCHITECTURE

Figure 2 is a functional block diagram of the DCS. External data enters a t the left. The data and clock are processed by a smoothing loop consisting of a phaselock loop (PLL) and first-in, first-out (FIFO) memory. The smoothing loop performs two functions. First, it removes any jitter whose phase-noise spec- trum is 1 Hz or greater from the input data and clock. This preprocessing is necessary when the data source is a wideband tape recorder, which may contain jitter that could degrade the performance of the receiver under test. Second, it allows the insertion of a time- varying offset in the FIFO delay to simulate TEC envelope delay.

The symbol clock frequency source depends on the mode. With external data input, the PLL locks to the average frequency of the external bit clock. In internal data mode, the clock's numerically-controlled oscillator (NCO) free runs, executing frequency commands from the parameter memory.

Next the data is formatted into an application-specific frame format, including framing and multiplexing. Rate-1/2 convolutional encoding is performed if selected.

The resulting digital I and Q data is oversampled a t a t least 2 times per symbol to prevent aliasing. The samples are applied to the FIR filters, which convolve the data stream with the current CIRF. There are six filters, for 3 diversity channels times two components (I and Q) per channel. The signal a t the output of each FIR filter is multiplied by a 16-bit digital signal gain and applied to a 12-bit digital-to-analog converter (DAC). The result is 3 channels of analog I and Q baseband exhibiting the channel fading effects.

21 MHz CLOCK d

COEFFICIENT TRANSFER 4

The analog I and Q signals are modulated onto a 160- MHz IF carrier in a linear vector modulator. The IF frequency is controlled by the carrier NCO, which receives frequency commands from the parameter memory. The modulator is followed by a SAW filter which models the satellite transmit filter.

The scintillated IF signals are sent to the upconverter, where they are corrupted by a noise level controllable from the parameter memory. The signals are then upconverted to the application-specific carrier frequency (e.g., S band) for output.

STATE MACHINE

CHANNEL CONVOLVER IMPLEMENTATION

The FIR filters perform a complex convolution (I and Q data with complex coefficients). The Inmos IMS A100 FIR filter chip [3] is used, resulting in an overall DCS throughput of 12 billion multiply/ accumulates per second in a total of 36 A100 chips (6 per FIR filter card). A unique architecture was used to minimize the FIR-chip count.

The A100 chip is designed to accept 16-bit data and 4- bit filter coefficients. However, since the data input to the DCS is binary (1 bit) and the channel weights require 16 bits of resolution to preserve dynamic range and fading fidelity, a better f i t and greater hardware efficiency were realized by operating the chip "back- wards." That is, the data uses the 4-bit coefficient port and the coefficients use the 16-bit data port. It was determined that this arrangement can produce a normal FIR filter output transfer function if the data and coefficients are properly buffered, permuted and sequenced a t the chip's inputs, as shown in Figure 3. Using this method, the FIR filters with 96 complex taps and 16-bit weights were implemented in a single 8" x 11" card for each I or Q channel, including the DAC for analog baseband output.

Figure 3. FIR Filter Block Diagram

29.1.3. 0659

Page 4: [IEEE IEEE Conference on Military Communications - Monterey, CA, USA (30 Sept.-3 Oct. 1990)] IEEE Conference on Military Communications - A frequency-selective, digitally implemented

EXAMPLE CHANNEL REALIZATIONS

Figures 4 through 7 are photographs of RF S-band spectra produced by the DCS. Figure 4 is a 2.5 Mbps BPSK signal with no channel perturbation (single active tap in the FIR filter) and without added noise. The sidelobes are filtered out by a SAW filter implemented a t IF, modeling the transmit filter. Figure 5 is the same spectrum with added noise (AWGN). Figure 6 is the same spectrum as in Figure 4 with frequency selective channel effects. The channel represents a single multipath reflection with magni- tude equal to the main path and zero phase shift. The FIR filter contains two active taps of equal magnitude and phase, separated by an integral number of symbols delay. Figure 7 is identical to Figure 6 except that the multipath reflection has a phase shift of 180 degrees relative to the main path. Observation of fast frequency selective fading with an ordinary laboratory

spectrum analyzer requires care since the rate of change of the channel can interact with the sweep rate of the analyzer. These photographs were taken with the channel frozen a t a constant CIRF.

ENHANCED-PERFORMANCE DCS

The primary limiting factor in simulation accuracy has been found to be the analog vector modulator. An enhanced-performance DCS currently under develop- ment will utilize all-digital modulation from baseband to the first IF, providing improved fidelity and repeatability in deep fades. Other enhancements include finer channel sampling (40 samples per decorrelation time), closed-loop calibration of C/No, and scenario storage on hard disk. A single-board 68030-based VMEbus computer will replace the PC controller for added performance.

Figure 4. BPSK signal with no channel perturbation and no added noise.

Figure 6. BPSK signal with single in-phase multipath reflection and no added noise.

Figure 5. BPSK signal with no channel perturbation with added noise.

Figure 7. BPSK signal with single out-of -phase multipath reflection and no added noise.

29.1.4. 0660

Page 5: [IEEE IEEE Conference on Military Communications - Monterey, CA, USA (30 Sept.-3 Oct. 1990)] IEEE Conference on Military Communications - A frequency-selective, digitally implemented

CONCLUSION REFERENCES

[l] B. Currivan, "An Adaptive-Equalizer-Based Receiver for QPSK Satellite Downlink Signals in Ionospheric Scintillation," MILCOM '88 Classified Conference Record, Volume 1, Pages 81-85, October, 1988.

[2] R. A. Dana, ACIRF User's Guide, MRC-R-1198, Mission Research Corporation, 15 December 1988.

[3] The Digital Signal Processing Databook, Inmos/SGS-Thomson, 1989.

This paper has presented the functional and implemen- tational approaches to a digital dispersive-channel simulator. The design utilizes digital time-varying FIR filters to implement the fading channel, allowing a sig- nificant delay spread and data rate to be implemented in a single chassis.

ACKNOWLEDGMENTS

Acknowlegment is given to the Satellite Terminal Operation team who developed the DCS.

29.1.5. 0661