Phil Sussman, KB8LUJ (Note 1)
Ken Hopper, N9VV
Julius Breit, W9UWE
Michael Meyer, DL1GBM


This is the complete text of our article.
We welcome your comments.
Send your email to - Thanks !

After reading "A Comparison of HF Digital Protocols" in July, 1996 QST, we found it generally to be a fine piece. A lot of time and hard work were involved and we congratulate the authors for their fine article. We noted, however, that some explanations were confusing and other important points were missed. This article is offered to better explain the results. We also seek to expand and clarify the original presentation.

When comparing modes we determine the relative 'code' strength and reconstruction ability by measuring throughput under a variety of conditions. Usually there are four specific cases: AWGN (Note 2), CCIR-Good, CCIR-Poor, and Rayleigh Fading (Note 3). Ease of operation, cost, software features, and hardware sophistication are subjective and are not factors to consider when making measurements.

All tests must be standardized and run under the same conditions. We rely on four common limits: scale, bandwidth, frequency stability, and compression.

SCALE: The graphs used in the original article had a variety of horizontal and vertical scales. For example, CLOVER was charted to +40db (Note 4) while other modes were not. Also the horizontal scale always started at 0 dB rather than a lower figure. By starting at -20 dB, low end, 'into the noise' performance can be investigated.

BANDWIDTH: Since most HF digital modes are designed for operation within a 500Hz bandwidth, it would be best to test them at that bandwidth. Unfortunately most modern propagation analyzers have a 4000-5000Hz bandwidth, so results must be judged accordingly. Phil Karns, KA9Q, (Note 5) believes more horsepower can be gleaned when spread spectrum techniques are applied to HF, but that is a matter for another article.

FREQUENCY STABILITY: Many hams still use older equipment which is prone to drift. Most modern HF rigs are very stable; but, they still drift a bit, too. For example, AMTOR, because of it's slower speed and simplified coding can tolerate drifts of over 100 Hz. PACTOR-1 can remain linked at about 60-80Hz. So, AMTOR may outperform PACTOR-1 on rigs that drift.

Phase shift keying (Note 6) used by CLOVER and PACTOR-2 requires a frequency stability of about 10Hz. PACTOR-2 has built in tracking, which automatically compensates for drift up to about 100Hz. Since the test equipment is far more stable than the average HF rig, the problem of drift is sometimes unfortunately overlooked.

COMPRESSION: Compression (or lack thereof) can skew test results. Since compression is inherent in G-TOR, PACTOR, and PACTOR-2, but external in CLOVER the authors are to be commended for recognizing that fact when making their comparisons. (Note 7)

However, it should be noted that on-line compression with run length encoding acts on data based upon content. Other modes that use precompression techniques will compress all data, rather than parts of it. In either case the file length is insignificant.

Huffman Compression provides an internal factor of about 1.5 and Pseudo-Markov (PMC of PACTOR-2) has a 1.9 internal compress factor. The CLOVER mode uses external compress. The original QST data takes this into account, so we offer no changes, only an explanation.


We have regraphed the original QST data with common scales and extended them from -20db to +40db by incorporating data from a study done in Switzerland. (Note 8) Figure #3 graphs all modes in CCIR-Good Conditions and Figure #4 graphs all modes for CCIR-Poor conditions. It amazed us how the QST data was usually within 5 percent of the Swiss data.

Ideal AWGN conditions are shown in Figure #1. (Note 2)

Fading is a significant factor on HF. Rayleigh fading (Note 3) results are shown in Figure #2. (Note 8) With a noise bandwidth of 4KHz, the doppler spread (Hz) determines the mean fading rate:



The original authors did not plan to identify modems, only the modes. But, there are disparities between different modems running the same mode, namely MODEM A vs MODEM B on PACTOR (Note 9), and PCI-4000 -vs- P-38 on CLOVER. (Note 8)

In Table 3 the baud rates indicated were effective, not actual, since compression was ALSO being measured.

There are some confusing points about automatic -vs- manual regarding CLOVER and PACTOR-2. CLOVER changes timing AND modulation scheme based on band conditions, while PACTOR-2 changes modulation scheme (ie. DQPSK, D8-PSK, etc) based upon throughput and the amount of data in the buffer. PACTOR-2 has a standard and an extended data frame but does NOT change overall CLOCK timing. Speed changes (100b/200b of PACTOR-1 and DBPSK/DQPSK/D8PSK/D16PSK of PACTOR-2) are ordered by the RECEIVING station. Neither CLOVER nor PACTOR do this manually.

Forcing a CLOVER modem into its fastest bias does NOT offer a fair presentation of CLOVER poor signal performance. Performance of CLOVER is also shown with the Swiss test results of the PCI-4000. (Note 8)

The statement that PSK 'FALLS FASTER' under CCIR-Poor is misleading. In CCIR-Poor conditions, ALL modes fall, but the data shows PSK does as well or better than FSK. PSK modes (CLOVER and PACTOR-2) are far more powerful than FSK in CCIR-Good conditions.

The ability to transfer large files and the ability to control the direction of data flow are separate items.

The FEMA throughput rates cited for PACTOR-1 in Table 5 are suspect because a unit lacking an A/D converter in the M-ARQ (MODEM B - Note 9) was used to make the tests. As noted by the authors this makes PACTOR comparisons difficult. Likewise the results for CLOVER depend upon whether a PCI-4000 or a P-38 is used. We have provided a graph of AWGN. (Note 2)


AMTOR has been fading into history. While a robust mode, it only has 5 bits and can not transfer ASCII. It does not effectively compete with the speed and error correction of more modern modes.

CLOVER is a PSK mode which uses a full duplex simulation. It is well suited for operation under CCIR-Good conditions, however, there are differences between CLOVER modems. A good, frequency stable, rig is required. Data is transferred automatically between two linked stations. Other ARQ digital modes requires the sending station to change the direction of data transmission (OVER the link) or the receiving station to interrupt and reverse the flow. (BREAK-IN)

G-TOR is an FSK mode that offers a high transfer rate under good conditions by operating at 300 baud with INCREASED BANDWIDTH. As noted in the QST article, we agree that 300 baud is a shortcoming. Indeed G-TOR performance falls drastically in CCIR-Poor conditions, as shown in Figure #3.

PACTOR-1 is a powerful FSK mode especially in CCIR-Poor conditions and is a standard on modern TNCs. However, the protocol is not uniform (Note 9) and some TNCs do PACTOR-1 much better than others.

PACTOR-2 is a robust and powerful PSK mode which operates effectively under both very good and very poor conditions. It uses strong logic and automatic frequency tracking. Both PACTOR-1 and PACTOR-2 use the same protocol handshake, making the modes compatible.

When overlaying the graphs we must compensate for compression when translating from bits/second to characters/second. Using characters per second can be confusing with AMTOR or when applying compression to ASCII. We are misled by the number of bits/byte which can range from 7.5 to 10 in plain text and varies wildly with compression. We could use a Nomograph to depict this ratio, but since ASCII is a 256 bit code, we assume a byte to be composed of 8 bits. (Note 7)

Under good conditions, the results show PSK (CLOVER and PACTOR-2) outperforms all FSK modes. We discovered it is far more important for ANY HF digital mode to be operated as linearly as possible and to have the levels properly adjusted. It is very easy to overdrive a rig and transmit a distorted waveform. We noted the best performance of all HF digital modes was BELOW ALC compression, at about 40 to 50 percent of full rated RF output power. Frequency stability is more important than RF output power and in fact HF digital modes actually operate quite well at low (5-10w) levels.

In conclusion, we believe that comparing HF digital modes is objective but selecting an HF digital modem is subjective. We therefore do not make any recommendations. Rather we leave it to you, the reader, to study the charts and make your own 'bang for the buck' decisions.

73 and thanks for reading.



(1) The writers:

(2) AWGN is Additive White Gaussian Noise or background noise. A graph of AWGN is assumed to be made under ideal conditions with only a slight amount of 'channel' noise.

(3) Rayleigh fading is a significant factor. W.T. Webb, L. Hanzo, R. Steele: "Bandwidth-Efficient QAM schemes for Rayleigh Fading Channels", IEE Proc. Pt I, Vol 138, No 3, June 1991, pp 169-175. See also.. Oliver Bister: "Communications of the Ionospheric Channel for Military Applications", http://www.alpes-; Also.. Weiwan Liu: "Error Performance of DPSK signals in Frequency selective Rayleigh Fading Channels", ASB9705, SFU/CA 20-AUG-93, Defences/930820.a.txt - Finally.. Yuze Zhang: "Optimal and Near- Optimal Joint Channel and Data Estimation for Rayleigh Fading Channels", speech at Purdue Univ, 15-AUG-95, MSEE-239A.

(4) A figure of +40db nears wireline quality. Depending upon the ratio (S/N -vs- C/N), results over a 500Hz channel (when measured at 4-5KHz bandwidth) approach ideal.

(5) Phil Karns, KA9Q, is well known in the digital community as a leading digital communications specialist. Email:

(6) Frequency-shift-keying (FSK) shifts between two known states. Phase-shift-keying (PSK) changes PHASE of a signal against some reference. Differential Phase-Shift-Keying (DPSK) changes PHASE with respect to succeeding pulses in the same data stream. FSK is send by either shifting a carrier frequency (F1B) or modulating SSB with two shifting audio tones (AFSK). When sending PSK, a complex audio waveform is transmitted by SSB. Tracking is much more critical for PSK, thus requiring more frequency stability.

(7) When making comparisons, compression can be an unknown variable. AMTOR has NO compression. CLOVER uses compression external from the mode, while PACTOR internally uses Huffman/PMC compression techniques. We assume a ratio of 8 bits/byte to create a uniform scale. By depicting a scale of bytes/second a measurement of compression is acknowledged with effectively higher baud rates.

(8) Expanded chart data was done by Schuemperlin Avionics Laboratory of Switzerland in April, 1995, using a Magnavox MX-518 S. This study used two CLOVER PCI-4000s (not P-38), two SCS PTC-2s, and two PACTOR-1s (MODEM A - Note 9). Where differences exist between the CLOVER PCI-4000 and P-38, both are shown. G-TOR was not extended, since no data was available from this study.

(9) Not all PACTOR-1 is the same. MODEM-A has an analog-to-digital converter in the Memory ARQ (M-ARQ), while MODEM B does not fully implement all PACTOR-1 specifications. See.. Phil Sussman: "Packet Perspective", QST, May 1996.

Click here to return to PACTOR NEWS HOME PAGE.

Copyright © 1996; 1998 - Phil Sussman - N8PS - All rights reserved.
Rev: 11-Jan-2009