IEEE Communications Magazine paper on G.hn

With the recent approval of G.hn at ITU last week, IEEE Communications Magazine couldn't have chosen a better timing for the publication this month of the paper "G.hn: The New ITU-T Home Networking Standard", by my friends Vladimir Oksman (Infineon Technologies) and Stefano Galli (Panasonic Corporation).

Oksman and Galli are some of the most active participants at the ITU group in which Recommendation G.9960 (the Physical layer of G.hn) was developed. The paper, which is available for download at the HomeGrid Forum website, provides a good overview of the G.hn standard, including the overall architecture, the Physical Layer (Recommendation G.9960) and the Data Link Layer (Recommendation G.9961).

The paper provides interesting information on the issue of the Forward Error Correction (FEC) used in G.hn.

LDPC vs CTC in G.hn

Discussions about the selection of FEC probably took more time than any other topic during 2008 and 2009 in G.hn. FEC is a very hot topic in any communications standard, because it has a huge impact on the architecture of chips that implement it. Dozens of papers on the advantages of Low-Density Parity Check (LDPC) codes vs Convolution Turbo Codes (CTC) were presented and discussed in G.hn. Ultimately, the group decided to adopt LDPC, due to its advantages in terms of performance and decoding complexity.

Comparing FEC codes is a difficult task, although simulations can be used to analyze their differences. The Figure above (quoted from this paper by Panasonic - ITU TIES account required) provides an easy way to compare them. Each line in the figure shows the performance of a different code. FEC Performance is usually measured in terms of "Block Error Rate" (BLER) vs "Signal-to-Noise Ratio" (SNR). A BLER of 10^-6 means that only 1 out of every million messages (10⁶) are received with error, while a BLER of 10^-3 means that only 1 out of every thousand messages (10³) are received with error.

Video applications such as IPTV require a BLER of between 10^-5 and 10^-6 or lower, to ensure that video quality is not impacted and that service users do not experience frequent artifacts in their TV screens. A higher BLER (for example, 10^-3) would cause frequent errors in end-users' screens.

Another factor that impacts performance is the number of iterations that a receiver can perform when decoding a message. Using more iterations increases the probability of decoding the message successfully, but it also increases power consumption and usually forces chips to operate at a higher frequency, which is undesirable. A paper by Galli (ITU TIES account required) presented to G.hn shows that, from the point of view of complexity, 4 CTC iterations are equivalent to 26 LDPC iterations (because each LDPC iteration requires fewer mathematical operations than a CTC iteration).

The Figure above can be read in two different ways.

1. Let's assume that a G.hn system is operating with a SNR of 1.9 dB. If the FEC is done with LDPC, and the number of LDPC iterations is kept to 26, then the system will experience a BLER of 10^-5. On the other hand, if CTC were used, with 4 CTC iterations, the BLER would be much higher (5x10^-4). In other words, the number of packets received with error would be 5000 times higher if CTC were used.
2. Let's now assume that a service provider wants to operate the system with a target BLER of 10^-5. If the system uses LDPC with 26 iterations, then the required SNR is 1.9 dB. On the other hand, if the system used CTC with 4 iterations, then the required SNR would have been 2.5 dB. In other words, the CTC system would require a power level 14.82% higher than if LDPC were used.

The conclusion is clear: G.hn systems using LDPC provide better performance than CTC systems (when operating at the same SNR level), or are capable of operating with lower power levels (when operating at the same BLER level).

The paper by Oksman and Galli also provides interesting information regarding the choice of OFDM parameters in G.hn, and shows how these are optimal from the point of view of performance, given the special characteristics of each G.hn media (power lines, phone lines and coaxial cables).

Comments

[VirtGirl]

It is really complex...

[Cheap Computers Canada]

Technicalities always make my head move in circles.