MIL-STD-188-110B and STANAG 4539

In the mid-1990s, researchers at the Communications Research Centre in Canada were conducting a field trial to characterize improvements that could be obtained using polarization diversity, both for transmit and receive operation [30]. The test signal for the trial was to be transmitted alternately on vertically and horizontally polarized antennas, while the signal was received and recorded on a distant pair of colocated vertically and horizontally polarized antennas. With this setup, it was possible to look at all combinations of transmit and receive polarization and, on receive, to directly compare receptions from either single antenna with those that combined the signal from both antennas in a diversity arrangement.

The test signal consisted of sequential transmissions of various data rates achieved with single-tone modulation, combined with FEC based on convolutional codes. The increase in available processing power (compared to 1980s DSP technology) meant that modified equalization techniques, capable of effectively using more efficient waveforms, were practical. In an attempt to characterize the diversity gain performance as fully as possible over a wide range of conditions, the PSK modulations used in the trial mapped to data rates from 75 bps BPSK with repetition coding to 4800 bps using a punctured 8PSK modulation.

Interestingly, signals based on QAM modulations were also included, which provided a series of data rates up to 9600 bps as part of the test signal.

The design intent of the test signal was to have a selection of signals which, at the low data rate end, were more robust than would ever be needed on the link, and on the high end, were more complex than could be supported by a skywave link. However, the results of those trials led to the recognition that the higher order QAM constellations were practical not only for use on surface-wave channels, or in conjunction with diversity combining, but in many cases worked well over skywave channels.

During the same time frame, the early drafts of NATO STANAG 5066 were being written. This new profile for HF data communications clearly stood to benefit from the provision of a new standardized waveform to achieve data rates greater than those offered by the existing STANAG 4285 and MIL-STD-188-110A serialtone waveforms. This led to the inclusion of Annex G in the draft STANAG 5066, which defined a waveform providing data rates of 3200, 4800, 6400, and 9600 bps, using Q PSK, 8 PSK, 16 QAM, and 64 QAM, respectively. The choice of data rates above 2400 bps was a decision to avoid conflict and interoperability concerns associated with providing the 2400 bps rate already standardized in other waveforms.

Sadly, in international standards efforts, things rarely go smoothly. The NATO groups responsible for standardization felt that the new waveform did not belong as an Annex to STANAG 5066, which was categorized as a protocol standard. Instead, they preferred a separate STANAG to define the waveform.

Competition to define the new NATO-standard high-speed waveform ensued between groups in France, Germany, and a cooperative effort by developers in Canada and the United States. It was this latter cooperative effort that spawned Appendix C of MIL-STD-188-110B. The new waveform used the constellations and much of the waveform structure developed for STANAG 5066 Annex G and added further enhancements: a 32QAM constellation providing an 8 kbps data rate; a tail-biting convolutional code; a block interleaver with six interleaving depths; and a preamble that supported a full autobaud feature.

The competition held by the NATO working group pitted three competing designs against one another,

with scientists from the Defence Evaluation and Research Agency (DERA) in the United Kingdom conducting the testing on behalf of NATO. The entry from France was based on an OFDM modulation scheme utilizing a cyclic prefix and pilot tones in conjunction with a turbo-code FEC scheme. The entry from Germany was based on a PSK/QAM architecture, with a symbol rate significantly higher than the 2400 baud used by previous serial-tone waveforms. The higher symbol rate resulted in some degradation in performance when the waveform was constrained by radio filters to fit within the usual 3-kHz channel allocation. The final entry in the competition was an implementation of MIL-STD-188-110B Appendix C developed by the Harris Corporation of Rochester, New York and Melbourne, Florida.

The results of the competition were unambiguous, with the new MIL-STD waveform emerging as the clear winner. The end result of this standardization process was that the two standards (STANAG 4539 and MIL-STD-188-110B Appendix C) adopted the same waveform. The performance specifications in the STANAG were slightly different, as they required the use of radio filters in the simulation environment, and the required performance targets in the STANAG were somewhat more difficult to meet, to ensure that STANAG modems reach the performance levels not achievable by the competing waveforms.

3.2.3.1 Overview and Waveform Structure

The characteristics of the STANAG 4539 and MIL-STD-188-110B Appendix C waveform are summarized in Table 3.4.

The waveform structure is shown in Figure 3.9. To facilitate transmit ALC and receive AGC functions, the transmitter may optionally send up to 7 repetitions of the complex conjugate of the first 184 symbols of the initial preamble. These first 184 symbols of the initial preamble are a pseudo-random BPSK modulated sequence, chosen for good correlation properties to aid signal detection. This is followed by a 103-symbol segment comprised of two 32-symbol segments, each containing a cyclic repetition of a length 16 Frank-Heimiller (FH) sequence, separated by three 13-symbol Barker sequences. The three Barker sequences are quadrature modulated to encode the data rate and interleaver settings. This allows the receiver to decode the transmission without a priori knowledge of the data rate and interleaver settings (the famous MIL-STD

Figure 3.9 Appendix C waveform structure.

The initial preamble is followed by alternating blocks of 256 data symbols and 31 known miniprobe symbols, where the miniprobe symbols are constructed from the same cyclically extended 16-symbol FH sequence.

The designers of the Appendix C waveform thought that it would be useful to include a regularly reinserted preamble. This regularly reinserted preamble is the same as the final 103 symbols of the initial preamble, and allows acquisition of the signal if the initial preamble is missed and also simplifies timing correction for long transmissions. The regularly reinserted preamble is the reason for the 31-symbol miniprobe length.

A waveform made up of blocks with 256 data symbols and 32 known symbols has a waveform efficiency of 8/9. When coupled with a symbol rate of 2400 bps and a rate-3/4 FEC code, this leads to desirable data rates for synchronous interfaces for most modulations. For example, with 6 bits per symbol, it results in a 9600-bps user data rate. However, periodically reinserting a preamble will reduce the effective efficiency of the waveform and hence throw off the data rate calculation. In the Appendix C waveform, the 103 symbols of the reinserted preamble may be thought of as being composed of 31 symbols of the FH sequence from the data block immediately preceding the reinserted preamble, plus 72 additional symbols.

The efficiency penalty for each of these additional symbols has been made up by using 31-symbol miniprobes (instead of 32 symbols) in each of the 72 blocks between reinserted preambles.

The FH-based miniprobe sequence that is transmitted can be one of two phases, positive (+) or inverted (–). Miniprobes are modulated, positive or inverted, with the autobaud information over the 72 blocks between reinserted preambles. This allows the receiver to infer the data rate and interleaver setting from only the miniprobes, without necessarily waiting for the reinserted preamble. In practice, this feature does not work effectively in real HF channels. In most cases, if the initial preamble is missed, signal acquisition will coincide with the reinserted preamble.

3.2.3.2 Modulation: PSK and QAM

The most innovative feature of the Appendix C waveform, when compared to previous HF waveforms, is its use of quadrature amplitude modulation (QAM) for data rates of 6400 bps and higher. 16-, 32- and 64-QAM constellations have been used. The choice of 64-QAM constellations (rather than PSK constellations) is mandated by the need to improve bandwidth efficiency in order to achieve higher data rates, and by the increasingly poor signal space distance properties of PSK constellations beyond 8-PSK. The 16-QAM and 64-QAM constellations are shown in Figure 3.10. A notable feature of these constellations (relative to classical square QAM constellations) is the way points fill the area within a circle of constant radius,

maximizing signal space distance, while minimizing peak to average ratios.

These constellations have been carefully designed and retain the desirable Gray-coding features of the square constellations. That is, for most of the constellation points, a decoding error that results in an incorrect selection of a nearest neighbor results in only a single bit error in the output.

Figure 3.10 HF QAM constellations.

3.2.3.3 Demodulation of Data Blocks

The data blocks of the Appendix C waveform consist of 256 symbols of data to be demodulated with cyclically extended 16-symbol FH sequences immediately before and immediately following the data block.

The cyclic FH sequences have perfect correlation properties, and as such, provide good channel estimates before and after the data block. These channel estimates are based solely on known data and do not depend in any way upon decisions made by the receiver. This is different from the equalization approach normally used with previous generation waveforms where an LMS update procedure incorporates decisions made on data symbols to update the channel estimate.

The length of the FH sequence limits the length of the channel impulse response estimation to 16 symbols, or 6.7 ms, when this approach is used. This results in a delay spread capability that is in line with 110A and STANAG 4285 waveforms. The Doppler spread capability is somewhat reduced, however.

While the provision of good channel estimates before and after the data block improves performance significantly, relative to having to use known data in the channel estimation process, the large interval between channel estimates results in somewhat poorer resistance to Doppler spread compared to the older, lower-data-rate waveforms.

In most cases, the increased waveform efficiency (8/9 for Appendix C versus 2/3 for 110A 2400 bps or STANAG 4285) results in better performance by the Appendix C variant because of the ability to use lower modulation complexity to achieve the same data rates. For example, the 3200-bps data rate in Appendix C provides better performance on most channels than the 2400-bps data rate in either MIL-STD-188-110A or STANAG 4285, as it uses a QPSK modulation instead of 8-PSK.

3.2.3.4 Forward Error Correction

The Appendix C waveform introduced a fully tail-biting convolutional code as part of the FEC scheme.

Tail-biting improves efficiency significantly in packet-based systems, as it eliminates the need to flush the encoder at the end of a transmission (as was done previously). With the new tail-biting FEC, it became possible to precisely match the data to be transmitted to the duration of the time slot available, an important feature for optimizing ARQ or slotted transmissions

The FEC scheme has additional features that benefit packet-based transmissions. Like the 110A waveform, the Appendix C waveform uses a block interleaver. However, a much wider range of interleaving options is provided, with 6 choices ranging from an ultrashort interleaver of 0.12 s to very long interleaver of 8.64 s. The very long interleaver provides good performance in fading for broadcast transmissions, while the shorter interleaver options allow users to make trade-offs between latency and resistance to fading effects.

In many applications, the cost of the long latency associated with the very long interleaver overshadows whatever gains result from the performance advantage in fading channels.