Rev C--January 8, 2003
1. Review of Terms of Reference
A survey conducted in 2000 by the erstwhile ARRL Technology Working Group indicated that digital voice was among the most popular potential technologies for Amateur Radio. At their July 2000 meeting, the ARRL Board of Directors voted to proceed with the development of digital voice. Shortly thereafter, the Digital Voice Working Group (DVWG) was formed and the Technology Task Force (TTF) set certain goals for us. We ourselves followed with certain more specific technical goals.
Goals
From the TTF's January 2001 report, recommended goals for the DVWG were:
The DVWG's initial technical
goals for digital voice systems included, but were not limited to, 1) audio
quality sufficient to become popular, 2) publication of protocols and
modulation formats so that others might reproduce systems and improve on them,
3) robust error detection and correction to provide reasonable performance in
the presence of noise, interference and dispersion, 4) simultaneous voice, data
and even image capabilities, and 5) a standard way of evaluating audio voice
quality. A general goal was to
investigate use of all proprietary and public standards already in place for
digital voice. Another was to reach out
to experimenters around the globe to identify and support planned and ongoing
efforts.
The main aims overall were
obviously to get working digital voice systems into the hands of radio amateurs
and to promote understanding of them through publications and forums. Well, that has happened; but a few issues
arose along the way.
Hindsight
One thing we did not anticipate was the short remaining life of the AMBE-1000 digital voice codec chip from Digital Voice Systems, Inc. (DVSI) and the Motorola MC56002, around which the G4GUO system was designed. By the time the system was ready for widespread testing, both parts were about to become obsolete.
Other microprocessors were there and we managed to get systems reconstructed, this time with modems in software. Those systems are only just now ready for initial testing but the AMBE-1000 is still dying. Meanwhile, a group of students at Temple University under Dennis Silage, K3DS, have begun a new design with the more modern AMBE-2020 chip. In at least one mode, the new chip is compatible with the old. Our work on a software modem still may apply to that system.
Another thing we might have seen coming was the rapid acceptance by regulatory bodies around the world of digital audio broadcasting standards. Their use of 10 kHz or more of spectrum, though, made them seem impractical for Amateur Radio. We became aware of a 3-kHz-BW derivative about a year ago.
We discovered that the IMBE technology from DVSI, used in APCO25, Inmarsat and Iridium is obtainable only through license, at significant cost. Our conversations with DVSI were amicable enough, but the company is obviously in business to make money and negotiations over special software licensing of IMBE were ultimately unfruitful. Some hams have obtained APCO25 radios and use them with compatible digital repeaters; but they are relatively few and far between-- they cost too much for most hams. AMBE is available in chip form; IMBE only as software.
2. Accomplishments
The DVWG obtained agreement from DVSI's President Dr. John C. Hardwick to make AMBE chips available to amateurs at reasonable cost and in small quantities. Large-quantity discounts are possible. DVSI posted a copy of Reference 1 to their Web site with ARRL's permission.
Our investigation of proprietary and public standards has been thorough. We have identified both roadblocks and open highways to further Amateur Radio development of digital voice. We have reached out to new friends in Europe, Asia and in the North America to learn about and test their systems.
In our roughly two and a half years, Group members and associates have published at least six articles1-5 about digital voice in ARRL periodicals, now five reports and various articles on the Web, and two chapters in different books. We conducted a public forum at Dayton 2002, the record of which is also published on the Web,6 and we plan a few more for 2003.
We adopted mean opinion score (MOS) as the best way of evaluating voice quality, digital or otherwise. We conducted extensive tests of digital voice systems, including APCO25, G4GUO, Alinco, AOR (ARD9800), ICOM (D-Star) and Thales. Some of these also have simultaneous data or video capabilities and include error-detection and -correction features. We find that hams will trade away some voice quality for such additional services. With an average MOS of 2.5, the Alinco system alone was found somewhat disappointing. Our goals for audio quality and robustness in the presence of noise and interference have been met by the other systems. A summary of test results is in Appendix A.
We discovered early that digital phone, like analog phone, belongs on the phone bands. We recommended that digital phone be announced as is slow-scan television-- in analog phone before and after transmissions-- until such time as it becomes widespread.
We have shown the technology to be practical by on-air demonstration of it on HF, VHF and microwaves, fixed and mobile. This Group set the distance record for an HF digital voice contact on November 22, 2002 between Sevierville, Tennessee and Paris, France in the 15-m band.
Finally, there are some inherent limitations to medium- and high-speed modulation methods for Amateur Radio. Those are noted in Appendix B.
3. Conclusions and Recommendations
We have stated all along that digital voice standards could only be established by actual usage. Just as many data communications formats exist in Amateur Radio, there appears to be room for more than one medium- or high-speed multimedia format. And just as the widespread acceptance of SSB took considerable time, so will the acceptance of digital voice.
Our work began at a time when many disparate efforts were already under way. In fact, all the formats listed in Appendix A existed as we started in 2000. Some were tooled for commercial applications and some for ham radio. The divergence that existed 30 months ago has narrowed somewhat; but we identify three systems that are very close to one another: G4GUO, AOR and the Temple project. For those interested in free and open standards, those are the closest. We have commitments from the parties to work toward at least one mode providing interoperability.
All the others are based on some public standard. The Thales system stands out as the one PC-based solution; but significant commercial interest exists for it and it remains to be seen whether its lack of a special hardware requirement will engender popularity.
We have found that the number of persons willing to devote large amounts of time and money to privately developing complex systems for a limited market is quite small. The number who can actually do it is even smaller. These are major designs involving state-of-the-art techniques that only two years ago were largely unexplored in Amateur Radio. Further, the interests of those who experiment for the sake of it and those who want to make a profit are often diametrically opposed.
In addition, regulatory issues are in play. Hybrid systems involving simultaneous voice and data (SVD) do not seem to have specific support in our rules. Further, the data capabilities of some systems use symbol rates in excess of 300 bauds. While other systems such as Pactor II, III and Clover in use now on the HF ham bands also exceed that rate, the rules may hold us back by providing ammunition for those opposed to certain strange signals appearing in unexpected places. That is perhaps the single most important issue impeding further progress.
Finally, we note that much of the work done on medium- and high-speed modulation formats for digital voice systems is directly applicable to data systems. We fully expect the release of those systems to make a significant positive impact on amateur use of digital modes.
Our recommendations therefore are these:
A) Modify band plans to provide for voice-bandwidth digital emissions of all kinds.
B) Support a rule interpretation that allows multimedia operation based on the primary content of emissions, and that interprets the 300-baud limit as pertaining to individual subcarriers, not to the total.
C) With IARU and others, discuss the implications of third-party traffic-- including digitized voice-- through radio-to-Internet gateways.
D) Assign to the High-Speed Multimedia Working Group any remaining questions on points A-C above.
E) Acknowledge as viable all the digital voice systems we found acceptable.
F) As the Digital Voice Working Group has achieved its goals, give it new and specific tasks or kindly dissolve it.
Respectfully yours in service,
Gary Barbour, AC4DL
George Bednekoff, AC5WO
Charles Brain, G4GUO
John Gibbs, KC7YXD
Jesse Morris, KC5GTK
Doug Smith, KF6DX, Chair
Mike Tracy, KC1SX, HQ Liaison
References:
1. C. Brain, G4GUO and A. Talbot, G4JNT, "Practical HF Digital Voice," QEX, May/Jun 2000.
2. D. Smith, KF6DX, "PTC: Perceptual Transform Coding... Part 1," QEX, May/Jun 2000; Part 2, Mar/Apr 2001.
3. D. Smith, "Digital Voice: The Next New Mode?" QST, Jan 2002.
4. D. Smith, "Digital Voice: An Update and Forecast," QST, Feb 2002.
5. C. Demeure and P. Laurent, "International Digital Audio Broadcasting Standards: Voice Coding and Amateur Radio Applications," QEX, Jan/Feb 2003.
6. Dayton 2002 Digital Voice Forum, images and audio, www.tapr.org/tapr/dv.
7. Technology
Working Group report to the Technology Task Force, Jan 2001.
Appendix A: Survey of Amateur Radio Digital Voice Systems
|
System |
MOS |
Modulation scheme |
Codec |
Raw data rate |
SVD? |
Error correction |
|
G4GUO |
3.3 |
OFDM/PSK |
AMBE-1000 |
3.6k |
no |
FEC |
|
Kachina 505RC |
3.8 |
FHSS/DSSS |
CVSD |
38.4k |
yes |
iterated coding |
|
D-Star |
3.4 |
GMSK |
CELP |
128k |
yes |
FEC |
|
Alinco |
2.5 |
GMSK |
CVSD |
14.4k |
no |
none |
|
AOR |
3.6 |
OFDM/PSK |
AMBE-2020 |
3.6k |
yes |
FEC |
|
APCO25 |
3.4 |
PSK/FSK |
IMBE |
9.6k |
yes |
FEC |
|
Thales |
3.4 |
OFDM/PSK/QAM |
MPEG-4/HSX/SBR |
3.2k |
yes |
FEC |
Glossary of acronyms:
AMBE: Advanced multi-band excitation coding
CELP: Code-excited linear predictive coding
CVSD: Continuously variable slope-delta modulation
DSSS: Direct-sequence spread spectrum
FEC: Forward error correction
FHSS: Frequency-hopping spread spectrum
GMSK: Gaussian minimum-shift keying
HSX: Harmonic stochastic excitation
IMBE: Improved multi-band excitation coding
MPEG: Moving picture experts group
OFDM: Orthogonal frequency-division multiplexing
QAM: Quadrature amplitude modulation
SBR: Spectral band replication
SVD: Simultaneous voice and data
Appendix B: OFDM Performance Limitations
OFDM Basics
To get information from one point to another via radio, we have but three physical properties of radio waves to exploit: amplitude, phase and frequency. Really, phase and frequency are related in that FM and PM are both forms of angle modulation.
A high-speed digital modulation scheme using a single modulated subcarrier tends to produce a large occupied bandwidth. Multiple subcarriers may be employed to reduce bandwidth while maintaining high throughput capacity. That is the general idea behind OFDM (orthogonal frequency-division multiplexing). The symbol (baud) rate of each subcarrier is chosen with respect to anticipated channel impulse responses.
In a typical 3-kHz bandwidth system, dozens of individual subcarriers may be used. Each is modulated digitally, usually using some form of phase-shift keying (PSK) or quadrature amplitude modulation (QAM). Each subcarrier bears a fraction of the total throughput burden. For instance, a system having 36 subcarriers, each modulated with quadrature PSK (QPSK) at 50 bauds, produces a raw throughput rate of (36)(2)(50)=3600 bps. The factor of two indicates that at each symbol time, two bits of information are conveyed by QPSK. The actual rate of useful throughput is usually reduced by the need for synchronization and error-detection and -correction. Those requirements use up some of the total capacity.
In such a system, the center frequencies of each subcarrier are spaced evenly. That is, the frequency difference between sub-channels is constant. The designer would use the entire available bandwidth, perhaps selecting a 75-Hz separation. Sub-channel centers in Hz might then be 300, 375, 450, 525, 600 and so on.
Crest Factor
One issue in OFDM systems has to do with peak-to-average ratio, usually called crest factor. It turns out that combining multiple subcarriers as described may result in a signal producing occasional peak values far above its average or RMS value. That is a disadvantage because it limits the total energy that may be transmitted in peak-limited transmitters. Every transmitter is, after all, peak-limited in some way.
Coded OFDM (COFDM) generally refers to a coding technique that limits the effects of certain channel impairments. COFDM has also been found useful in reducing crest factor by exploiting redundancy among sub-channels.1 When used in that way, however, the correction power of the coding is diminished. Some systems instead have employed simple clipping to reduce crest factor with some impact on other performance factors.
Intermodulation Distortion
In both transmitters and receivers, intermodulation distortion (IMD) forms a distinct limitation for OFDM systems. Because of cube-law nonlinearities, any two adjacent subcarriers produce IMD products by their very presence that fall near the center of adjacent sub-channels-- right where they do the most damage. Two or more nonadjacent subcarriers may or may not produce IMD that also falls in another sub-channel.
QAM involves both amplitude and phase modulation. Each QAM subcarrier therefore produces IMD on its own because its amplitude is continually changing.
IMD shows up as noise in demodulated data from any particular sub-channel, since the data in one sub-channel usually do not correlate with those in another sub-channel. Such IMD shows up as a "noise cloud" around the points in a constellation display, often used to show coherence of received signals, and in signal-to-noise ratio (SNR) calculations performed on those signals.
Using OFDM signals, typical Amateur Radio HF transceivers employing class-AB power amplifiers (PAs) produce recovered SNRs in the range 10-20 dB because of IMD effects. That, in turn, limits the bit error rates (BER) of OFDM systems. Automatic level control (ALC) may also produce deleterious effects.
The solution involves ensuring the highest degree of linearity possible in radio transceivers. That is most often achieved at power levels well under the maximum rating of transmitters. Class-A PAs may be a distinct advantage here. Digital signal processing (DSP) has been used to adaptively correct for linearity errors.
Pre-distortion of exciter waveforms is a useful method for combating IMD; however, protagonists have found that bandwidths of up to five times the corrected bandwidth must be used for pre-distortion. While the technique has shown success in reducing low-order IMD, it often increases high-order IMD. Post-correction filtering is therefore often required. IMD in receivers may similarly be corrected.
Pre-distortion requires knowledge of both amplitude and phase responses. Most solutions are narrow-band; thus, it may be that successful Amateur Radio implementations involve single-band operation.
Phase and Group-Delay Distortion
Where digital angle modulation is involved, the relative phases of signals are critical. When phase shift is not directly proportional to frequency, group delay is not constant. That may mean high frequencies propagate through transmitter and receiver faster than low frequencies, or vice versa. Data in one sub-channel may no longer correlate with those in another and some of the bits get out of time alignment, obviously degrading performance.
Adaptive DSP methods deal with group-delay distortion quite nicely by equalizing the channel to achieve relatively flat group-delay characteristics.2 Adaptive equalizers may use a training sequence or may adjust themselves "on the fly" to do their jobs. Such subsystems have been employed in telephone modems for many years and they have found their way into digital voice and other medium- and high-speed radio systems. Group delay in the transceiver itself, however, often limits the tolerable RF path characteristics because the total allowable spread is fixed by the capabilities of the equalizer.
Equalizers that continually adjust themselves are to be preferred over those using training sequences because data loss is minimized. Group delays that are continuously changing require that approach. Such is the situation on an HF radio path-- quite unlike the typical telephone line, which does not normally change much over short time frames. Short-term phase instability, though, remains an issue.
Phase Noise
Noise is the perennial enemy of the communicator. It is evidently present at all frequencies and it confuses digital demodulators according to well-known models. One source of noise over which digital communications systems designers have some control is called phase noise.
Phase noise is the unwanted phase modulation of frequency-control elements in a transceiver by thermal noise.3 Thermal noise is associated with the random (Brownian) motion of atomic and subatomic particles caused by heat. Many modern frequency-synthesis techniques have made great strides in reducing the phase noise of frequency sources, but such noise is still in play for the vast majority of systems.
In the case of closely spaced subcarriers in an OFDM system, each subcarrier mixes with a local oscillator (LO) in a transmitter to produce RF. In so doing, it carries with it the phase noise of the LO. That noise resides adjacent in frequency to each subcarrier in the transmitted wave. Some of it overlaps adjacent and nearby sub-channels, degrading their maximum SNR. LO phase noise in a receiver worsens the problem because it is uncorrelated with that in the transmitter and uncorrelated noise powers add.
When subcarriers are spaced as closely as 75 Hz, the effect can be quite significant. 50- or 60-Hz power-line frequencies may be present. Careful design and measurement are necessary to mitigate the effects of phase noise.
Interference
The presence of uncorrelated signals obviously degrades the performance of high-speed digital communications systems. Interference is often beyond the control of the operator. Determination of the interference immunity of any particular modulation format is a complex session in engineering statistics. Any theory about performance must be supported by empirical evidence.
Amateur Radio digital voice systems are currently allowed in the U.S. only on the phone bands. Interference thus is limited to that from other analog or digital phone or image signals. Digital voice systems tested to date seem to tolerate slightly less interference than their analog counterparts. On the other hand, when interference is mild, the effects of interference are reduced or eliminated by error correction.
Error Detection and Correction in Digital Systems
In the face of the above-mentioned obstacles, error detection and correction is strictly necessary for high-speed digital communications over radio. An OFDM signal transmitted with a fairly low SNR without error correction cannot stand much degradation on its way to a receiver and its demodulator before being hopelessly corrupted.
Many advanced algorithms have been developed that produce robust performance with minimal overhead. Overhead refers to the additional data that must be sent to achieve error detection and correction. The most rugged and efficient algorithms increase transmitted data by about 50%.
Acknowledgements
Many thanks to the following individuals for their helpful comments and suggestions on this article: Cédric Demeure, Thales Communications; Ulrich Rohde, Synergy Microwave; Charles Brain and George Bednekoff, ARRL Digital Voice Working Group-- Doug Smith, KF6DX.
References
B1. G. A. Davidson, "Digital Audio Coding: Dolby AC-3," in The Digital Signal Processing Handbook, V. K. Madisetti and D. B. Williams, Eds., 1998, CRC Press LLC, Boca Raton, FL.
B2. D. Smith, KF6DX, "Deconvolution in Communications Systems," QEX/Communications Quarterly, Sep/Oct 2001, ARRL, Newington, CT; also appears in 20th ARRL and TAPR Digital Communications Conference Proceedings, Sept 2001, ARRL.
B3. D. Stockton, GM4ZNX, "AC/RF Sources (Oscillators and Synthesizers)," in The ARRL Handbook for Radio Amateurs, 79th Ed.; D. G. Reed, Ed., 2001, ARRL.