Achieving Higher Data Rates

In Section 6.1, we described the current bandwidth allocations and regulations governing HF. In Section 6.2, we discussed applications that could benefit from higher data rates. Now let us investigate possible approaches for achieving higher user data rates.

There exist at least three possible approaches: the first is to increase data rates utilizing current 3-kHz allocations; the second is to increase data rates by using multiple 3-kHz contiguous or noncontiguous channels; and the third is by using wider contiguous bandwidth waveforms.

For this discussion, we will use a measurement related to SNR: the signal power-to-noise density ratio (SPNDR). SPNDR is the ratio of the total signal power to the noise power contained in a 1-Hz bandwidth;

it can be very helpful when comparing waveforms with different noise bandwidths.

6.3.1 3-kHz Waveforms

Increasing the data rate of a waveform in a fixed 3-kHz bandwidth is quite straightforward. The only variables available are FEC and the modulation density. Recall from Chapter 3 that M-PSK constellations above 8-PSK become very power-inefficient. More power-efficient M-QAM constellations are better choices for increasing data rates. Every time an M-QAM constellation size is doubled, there is approximately a 3-dB cost in SNR, while the information contained in the constellation increases by a single bit. As a bandwidth efficient modulation, M-QAM follows the Shannon capacity curve in requiring exponentially more power to achieve linear increases in data throughput.

Table 6.1 shows the required SNR and SPNDR, the constellation, and the data rates achievable for an AWGN channel (for a BER = 10^-4) using the MDR waveforms discussed in Chapter 3 and extrapolating to 19,200 bps (i.e., doubling the current data rate of 9600 bps while using the same FEC, which is a rate 3/4 code).

As can be seen in Table 6.1, the cost of doubling the data rate from 9600 to 19200 is 18 dB. In addition to this high SNR price, both the required dynamic range of the entire system and the peak-to-average ratio (PAR) of the waveform increase as the constellation size increases, creating an even greater overall cost. Table 6.2 provides a sampling of the worst case PAR for the MDR waveforms. Furthermore, the higher order constellations are even more susceptible to multipath and fading, which makes it extremely unlikely that the 4096-QAM constellation would be able to handle anywhere near the same multipath and fading channel conditions as 64-QAM. Although other choices exist for achieving 19,200 bps without requiring 4096-QAM and such high SNRs, reasoning based on Shannon’s capacity theorem suggest that the costs would similarly be very high. Undoubtedly, not allowing the bandwidth to increase beyond 3 kHz poses a significant challenge to the goal of realistically achieving higher data rates over HF.

Table 6.1 Data Rate, Constellation Size, Required SNR, and SPNDR for

3-kHz Waveforms Table 6.2 Worst-Case PAR for MDR Waveforms

Constellation PAR Before Radio Filter (dB) PAR After Radio Filter (dB)

4-PSK 2.4 4.9

16-QAM 4.0 5.7

64-QAM 5.1 6.6

In the last 10 years, there has been an enormous interest in multiple-input multiple-output (MIMO) systems as a way of increasing the data rates in a fixed bandwidth [5]. MIMO systems are also known as space-time modulation because of the added dimension of space that is created by the multiple antennas.

The basic concept is to transmit the data over multiple transmit (TX) antennas and to receive the data using multiple receive (RX) antennas. Note that the receiver will require special MIMO processing to demodulate all the transmitted/received signal pairs simultaneously. If the number of TX antennas is N and the number of RX antennas is greater than or equal to N, the data rate of the system can be increased by the value N (assuming there is enough multipath in the system to support the N channels [6]). This technique has mostly been applied to systems operating in the 2 GHz or higher frequency range (where antenna spacing is small due to the shorter wavelengths).

In HF, MIMO techniques may be difficult to apply because of the longer wavelengths, which result in antennas that must be large in size. When spatial separation is used to decorrelate signals, this will require larger separation between all the TX antennas and all the RX antennas (although an argument can be made for using collocated antennas with different polarizations in this context).

In addition, the “enough multipath” constraint required in order to achieve the N-fold increase in data rate may not be available. On an AWGN channel, MIMO systems do not provide opportunities for increasing data rates (as described above) because a multipath rich environment is required [7]. In any case, it seems clear that, at best, MIMO would likely only be applicable to communications between a relatively small subset of HF sites with sufficient real-estate in order to be able to support multiple large HF antennas.

6.3.2 Multichannel Waveforms

An approach for increasing data rates over HF that fits nicely with current bandwidth allocations and existing radio equipment is the multichannel approach. The idea behind this approach is simply to use multiple 3-kHz channels in parallel. This approach offers users a linear increase in data rate as a function of the number of channels available. Modem implementers face only a linear increase in computational complexity for demodulating the multiple channels.

Let us illustrate this approach by comparing the performance of the 9600 bps waveform of MIL-STD-188-110B Appendix C [13] (labeled 110B/C) to the 9600-bps independent sideband (ISB) waveform (using 4800 bps per 3-kHz channel), which is defined in US MIL-STD-188-110B Appendix F [13]

(labeled 110B/F). In order to compare the two approaches fairly, the total transmit power of both waveforms must be the same. In addition, since peak-power limited amplifiers (i.e., linear amplifiers) are used, the difference in the PAR of the waveforms must also be accounted for. For example, the SNR required by 110B/C to achieve a BER of 10^-4 on the AWGN channel is 19 dB. For the 110B/F waveform, the SNR required per channel is 11 dB. This yields a gross advantage of 8 dB for using 110B/F instead of 110B/C. If two separate radios are used to transmit 110B/F and the total TX power is required to be the same, the advantage of 110B/F drops by 3 dB to 5 dB. However, accounting for the difference in PAR (64-QAM PAR is 6.6 dB, 4800 PAR is 4.9 dB), the advantage of 110B/F increases to 6.7 dB. If a single ISB radio is used instead of two radios, the PAR of 110B/F increases by 2 dB. This increase in PAR happens when both sidebands are combined before power amplification (note that the 2 dB value is an actual measured value); the advantage of 110B/F in an ISB radio is now 4.7 dB. If more than two sidebands are combined, PAR continues to increase (i.e., eight adjacent sidebands in a single radio would increase PAR by about 8 dB). Table 6.3 compares the SNR required for each waveform to achieve a BER of 10^-4 on the three test channels of STANAG 4539. Even at 9600 bps, the benefits of the multichannel approach are significant.

An important assumption that was made for the 110B/F waveform in the previous results is that each channel was assumed to have the same average HF channel conditions and that the fading observed on each channel was independent of the other channel. This assumption may not be realistic for ISB channels because the fading observed on adjacent 3 kHz channels may not be completely uncorrelated. Furthermore, if the two channels are separated by 1 MHz (or more), the probability that both channels exhibit the same amount of multipath, fading, and SNR is very small. This assumption is the main drawback of US MIL-STD-188-110B Appendix F and other approaches like it [8] for obtaining higher data rates. These waveforms always use the same symbol constellation (i.e., 8-PSK, 16-QAM, etc) on all the available channels. They also spread the FEC and interleaving between all the channels. This requires that similar HF channel characteristics be observed on all channels for the waveform to function well. The system must discard channels that do not support a higher data rate available on other channel(s). In the worst case, this might result in only a single channel able to support the transmission with a good signal to noise ratio.

Other alternatives that could provide a more effective use of the multiple channels are: (1) develop a multichannel automatic repeat request (ARQ) protocol that attempts to maximize the data rate of each individual channel and can thus achieve the highest possible multichannel data rate utilizing standard modems, waveforms and radio equipment; and (2) develop a multichannel STANAG 4538 (3G) ARQ protocol.

Table 6.3 Comparison of 110C/C and 110C/F 9600 bps Waveform

The practical drawbacks to using the multichannel approach in the field are that many 3 -Hz HF channel allocations are required in addition to many radios, antennas, modems, and so on. Although in recent years there has been much discussion of multichannel radios, the authors are not aware of any current developments of HF radios with more than two channels (except for strategic systems where the intent is to use each channel for a different application rather than to use all channels for a single application).

A final open question on the topic of multichannel waveforms is whether users are ready to pay the high price required to implement multichannel waveforms. Although the 110B/F waveform is being deployed in naval applications, due in large part to existing ISB frequency allocations and ISB radio equipment, it is unclear whether or not users that only have SSB equipment are willing to tie up a large portion of their radio assets on a single high data rate link.

6.3.3 Wider Contiguous Bandwidth Waveforms

Channel bandwidth plays a crucial role in the design and performance of high data rate waveforms. As was presented in Section 6.3.2, a two-channel ISB waveform (6 kHz) can have a significant performance advantage over a single-channel waveform (3 kHz) when providing a data rate of 9600 bps. As the data rate is decreased, this advantage becomes smaller. However, as the desired data rate increases beyond 9600 bps, this advantage can be even greater. Table 6.4 compares the required SPNDR for achieving a BER of 10^-4 on a mid-latitude disturbed [9] channel for various data rates as a function of bandwidth for 110C/C waveforms and for a family of trellis-coded OFDM waveforms, as presented in [10]. Note that the PAR is not taken into account in this table, but would further increase the advantage of the wider bandwidth waveforms. What is astonishing about this comparison is that almost the same SPNDR is required to support 9600 bps in 3 kHz as is required to support 64,000 bps in 80 kHz. Unmistakably, allowing the bandwidth of the waveform to increase yields much more power efficient waveforms.

Table 6.4 Required SPNDR to Achieve a BER of 10^-4 for ITU Mid-Latitude Disturbed Channel

Data Rate (bps) Bandwidth (kHz) SPNDR (dB)

Of the three choices presented for increasing data rates over HF, the wider contiguous bandwidth approach offers the best performance. The PAR of the wider contiguous bandwidth waveforms would be the same as for 3-kHz waveforms; a multichannel approach implemented in a single radio would suffer from increased PAR due to the combining of the channels before the power amplifier. We therefore prefer the wider contiguous bandwidth approach.

The next decision is whether to use a single-carrier or OFDM waveform. As mentioned in Chapter 3, a possible drawback of single-carrier waveforms is the computational complexity of the adaptive equalizer. If we are to maintain the same delay spread capability, the number of taps required by the equalizer grows linearly with the increase in bandwidth, while equalizer complexity grows in proportion to the square of the number of taps. Although the computational complexity of the equalizer is a challenge for wider bandwidth waveforms, we have found that current digital signal processors (DSPs)—in combination with field-programmable gate arrays (FPGAs)—allow the implementation of single-carrier waveforms for bandwidths of at least 24 kHz. Considering the other drawbacks of OFDM waveforms (see Chapter 3), we prefer the

single-carrier waveform as long as the equalizer remains feasible.