Wp-cdm625 Acm White Paper
Wp-cdm625 Acm White Paper
Wp-cdm625 Acm White Paper
com
Adaptive Coding and Modulation (ACM) in the CDM-625 Advanced Satellite Modem
Author: Richard Miller Senior Vice President New Technology
June 2009
June 2009
Overview
Adaptive Coding and Modulation (ACM) is a very significant new feature in the CDM-625 Advanced Satellite Modem. This feature requires Firmware Version 1.4.0, or later, with a VersaFECTM plug-in module, and the appropriate FAST codes. VersaFEC ACM is a patents-pending technology, wholly owned and developed by Comtech EF Data and Comtech AHA.
ACM turns fade margin into increased link capacity gains of 100% or more are possible, compared to traditional Constant Coding and Modulation (CCM). This is accomplished by automatically adapting the modulation type and FEC code rate to give highest possible throughput. ACM maximizes throughput regardless of link conditions (noise or other impairments, clear sky, rain fade, etc). Initial set-up is easy, and then requires no further user intervention. With a CCM system, severe rain fading can cause the total loss of the link, and zero throughput. ACM keeps the link up (with lower throughput) and can yield much higher system availability. ACM in the CDM-625 is used in conjunction with VersaFEC and is currently for IP traffic only.
Background
ACM is not a new concept. It has been used for some time in wireless communications, including terrestrial microwave applications, and more recently over satellite links. Its primary function is to optimize throughput in a wireless data link, by adapting the modulation order used and the Forward Error Correction code rate (which both directly affect spectral efficiency, expressed in bits per second per Hertz), according to the noise conditions (or other impairments) on the link. Implicit in this concept is that the symbol rate (and power) of the wireless communication system must remain constant. This ensures that the bandwidth allocated for a particular link is never exceeded. Given that the symbol rate does not change, if modulation and coding are changed, the data rate must therefore be modified. This is expressed in the simple equation Symbol rate = bit rate / (modulation order * code rate) For example, for Rate 3/4 QPSK (where modulation order = 2) Symbol rate = bit rate * 0.666 Re-arranging: Bit rate = symbol rate * modulation order * code rate So, in changing to a higher modulation order or code rate, the bit rate is increased, and in changing to a lower modulation order or code rate, the bit rate is reduced. However, there are several important factors to be considered, namely:
The digital communications system must be able to tolerate a change in bit rate. Synchronous serial interfaces (such as G.703 E1, which operated at a fixed data rate of 2.048 Mbps) are totally unsuitable in a scheme where data rate is changing. The only 2|Page
June 2009
practical application for this scheme is a packet-based scheme that will tolerate a change in data rate, and which has mechanisms within its protocols to recognize when increased or reduced bandwidth is available. The best example of this is Ethernet, and this discussion is limited to schemes that employ it.
The bit rate cannot be changed arbitrarily - the link noise conditions - described in terms of Eb/No or SNR, must be able to support reliable communications for the given modulation order and code rate. This is a key point, as in fact, the link SNR is the input that drives the adaptation.
3|Page
June 2009
June 2009
Disadvantages of DVB-S2
While the scheme defined by DVB-S2 is undoubtedly very effective for many broadcast and higher data rate applications, it is definitely not a one size fits all solution. Here are some of the disadvantages:
Excessive latency the so-called short blocks are too long for low latency IP applications at low data rates. This is exacerbated by the addition of interleaving. It is overly complex in its implementation - the design of DVB-S2 dictated that all FEC blocks should be constant in bits. This means that for each ModCod, there are a varying number of symbols. This then makes the task of synchronization a much more demanding task. Also, because of the limitations of tracking the higher-order modulations in a very low SNR environment, so-called pilot symbols were added in order to aid tracking. Since the introduction of the original LDPC/BCH scheme an enormous amount of research has been done on the design of LDPC codes. The concatenated BCH code was added to mitigate the problem of error rate flaring and flooring. This is no longer necessary. Most importantly, however, LDPC codes can now be designed that yield almost equivalent coding gain, but with considerably shorter block lengths. In an ACM mode, no overhead channel was defined by DVB-S2 for the purpose of reporting SNR metrics to the originating end. It has been left to individual equipment manufacturers to decide their own methods. This illustrates that all ACM systems, DVB-S2 or not, are proprietary. In addition, it implies that additional bandwidth needs to be consumed for the SNR reporting, and this is not accounted for in the code rate.
VersaFEC ACM
VersaFECTM (a trademark of Comtech AHA), in concert with a novel ACM approach, addresses all of the shortcomings of DVB-S2 outlined above. There are patents pending for both VersaFEC and the ACM scheme. VersaFEC covers a family of 12 short-block LDPC ModCods, specifically designed for low latency and ACM applications. However, the VersaFEC codes are equally well suited to Constant Coding and Modulation (CCM) applications. The requirements for an ACM system that approaches the minimum possible latency are: The shortest possible LDPC codes that give performance at or very close to DVB-S2, in order to minimize latency, and which do no use interleaving. Design of the encoder to further reduce latency to the minimum possible. A constant number of symbols per block, to reduce the demodulator and decoder complexity, and significantly also reduces latency in the ACM case. The elimination of the need for pilot symbols for carrier tracking at low SNR by substitution of other modulation techniques. This further reduces the complexity of the demodulator. A reduction in the number of ModCods to further reduce complexity. 5|Page
June 2009
The inclusion, at the physical layer, of an overhead channel to permit the reporting of SNR metrics back to the originating end. Note that this does not have to be enabled or disabled it is part of the fundamental frame structure of VersaFEC ACM, and has been take into account in the code rate.
The family of VersaFEC short-block LDPC codes is presented below in Table 1. The modulation types include BPSK, QPSK, 8-QAM and 16-QAM. It will be seen from the table that in order to maintain a constant number of symbols per block, the block size in bits (data + parity) must necessarily change, depending on both the modulation type (which affects the number of bits per symbol) and the code rate. For VersaFEC, the block size varies between 2k and 8.2k bits. At worst, therefore, the VersaFEC codes are 50% shorter than the short DVB-S2 codes.
Modulation BPSK QPSK QPSK QPSK QPSK 8-QAM 8-QAM 8-QAM 16-QAM 16-QAM 16-QAM 16-QAM
Code Rate 0.488 0.533 0.631 0.706 0.803 0.642 0.711 0.780 0.731 0.780 0.829 0.853
Spectral efficiency, bps/Hz 0.49 1.07 1.26 1.41 1.61 1.93 2.13 2.34 2.93 3.12 3.32 3.41
Block size, bits 2k 4.1k 4.1k 4.1k 4.1k 6.1k 6.1k 6.1k 8.2k 8.2k 8.2k 8.2k
Typical Eb/No, for BER = 5 x 10-8 2.4 dB 2.2 dB 2.7 dB 3.4 dB 3.8 dB 4.6 dB 5.2 dB 5.6 dB 6.3 dB 7.0 dB 7.5 dB 8.0 dB
Min. Data Rate, CCM mode 18 kbps 20 kbps 23 kbps 26 kbps 28 kbps 35 kbps 39 kbps 43 kbps 53 kbps 57 kbps 60 kbps 62 kbps
Max. Data Rate, CCM mode 5.7 Mbps 10 Mbps 10 Mbps 10 Mbps 12 Mbps 12 Mbps 12 Mbps 12 Mbps 12 Mbps 14 Mbps 14 Mbps 16 Mbps
The VersaFEC codes compared with the Shannon bound are shown in the Figure below. It can be seen that the performance of VersaFEC at or near the DVB-S2 performance with 16 kbit blocks.
6|Page
June 2009
VersaFEC Codes
16, 0.829
16, 0.731
16, 0.780
QPSK capacity
Q, 0.803
1.5
Q, 0.706 Q, 0.631
Q, 0.533
B, 0.488
0 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14
SNR (dB)
Figure 2. VersaFEC versus constrained capacity Note that SNR is used in place of Eb/No - a convention for comparing ACM ModCods. SNR is defined as Eb/No + 10log(Spectral Efficiency).
7|Page
June 2009
VersaFEC implementation of ACM example case of 100 ksymbols/sec Spectral efficiency, bps/Hz 0.49 1.07 1.26 1.41 1.61 1.93 2.13 2.34 2.93 3.12 3.32 3.41 Bit rate (throughput) 49 kbps 107 kbps 126 kbps 141 kbps 161 kbps 193 kbps 213 kbps 234 kbps 293 kbps 312 kbps 332 kbps 341 kbps Minimum Latency, milliseconds, for each ModCod 34 + WAN BUFFER 32 + WAN BUFFER 30 + WAN BUFFER 28 + WAN BUFFER 26 + WAN BUFFER 30 + WAN BUFFER 28 + WAN BUFFER 27 + WAN BUFFER 27 + WAN BUFFER 26 + WAN BUFFER 25 + WAN BUFFER 25 + WAN BUFFER
ModCod 0 1 2 3 4 5 6 7 8 9 10 11
Modulation BPSK QPSK QPSK QPSK QPSK 8-QAM 8-QAM 8-QAM 16-QAM 16-QAM 16-QAM 16-QAM
Code Rate 0.488 0.533 0.631 0.706 0.803 0.642 0.711 0.780 0.731 0.780 0.829 0.853
OVERALL SYSTEM LATENCY = Worst-case ModCod (ModCod0) Latency = 34 milliseconds + WAN Buffer delay
By way of comparison, consider the same 100 ksymbols/second rate, but this time using DVB-S2. It becomes clear that there is an unintended penalty to having a constant number of bits per block (besides demodulator complexity). Each time the ModCod is lowered, and the throughput is reduced, the latency grows accordingly, due to the block size being related to data rate, not symbol rate. Remembering that, for the ACM case, the system latency is equal to the latency of the worst-case ModCod, DVB-S2 shows a severe penalty. For 16k short blocks, this calculates to be 329 milliseconds (+ WAN Buffer delay) versus 34 milliseconds (+ WAN Buffer delay) for VersaFEC ACM. For 64k block DVB-S2, the core latency is 4 times higher. Assuming a WAN Buffer of 20 milliseconds: Latency for 64k block DVB-S2 ACM at 100 ksps = 1336 milliseconds Latency for 16k block DVB-S2 ACM at 100 ksps = 349 milliseconds Latency for VersaFEC ACM at 100 ksps = 54 milliseconds
Summary: For the example shown, the latency for a 16k block DVB-S2 ACM scheme is approximately 7 times higher than VersaFEC ACM. The latency for a 64k block DVB-S2 ACM scheme is approximately 25 times higher than VersaFEC ACM.
8|Page
June 2009
IMPORTANT
Configuration is very straightforward. From the front panel, proceed as follows: 1) Under CONFIG, MODE, select IP-ACM. 2) Under CONFIG, TX, SYMB, enter the desired transmit symbol rate. Note that is a fundamental departure from the way in which the modem is configured traditionally. The Transmit symbol rate is limited by the FAST code installed. There are three options: 37 ksps to 300 ksps 37 ksps to 1200 ksps 37 ksps to 4100 ksps
IMPORTANT
3) Under CONFIG, TX, POWER, enter the desired transmit output level. 4) Under CONFIG, RX, SYMB, enter the desired receive symbol rate. Note that asymmetric operation is supported - transmit and receive symbol rates do not have to be equal. 5) Under CONFIG, ACM, configure the desired ACM operating parameters: a) Min/Max ModCod Here the user defines the range of ModCods that the system will operate over. ModCod0 is BPSK Rate 0.488 (0.49 bps/Hz), while ModCod11 is 16-QAM Rate 0.853 (3.41 bps/Hz).
If the user wishes to constrain the system to run at a fixed ModCod, set the Min and Max ModCod values to be equal.
IMPORTANT
The value of Max ModCod may be limited by other FAST codes installed. For example, suppose the 4100 ksps FAST option is installed, and the symbol rate set to 4100 ksps, the theoretical maximum data rate would be 14 Mbps at ModCod 11. However, if CnC is being used, with a 10 Mbps FAST limit, the Max ModCod will be limited to ModCod 7, or 9.6 Mbps.
WARNING
b) Unlock Action Here the user decides the desired action when the remote demod loses lock. This is important, as the ACM system depends on the feedback of the SNR metric from the remote demod to determine the optimum ModCod. The choices are: Maintain the current ModCod Go to minimum ModCod (recommended)
9|Page
June 2009
c) Target Eb/No margin This is a VERY important parameter. The ACM system is designed to switch based on thresholds that correspond to a BER of 5 x 10-8 for each ModCod. However, in order to prevent oscillation around two ModCods at this exact value, 0.3 dB of hysteresis has been added. The switch points and the hysteresis are shown below:
Figure 3.
The graph shows the switch points with the Target Eb/No margin set to 0dB. However, the switch points can be moved (increased) by configuring the Target Eb/No margin parameter, which can vary from 0 to 4.5 dB, in 0.5dB steps. In a fading environment it is highly recommended to add sufficient margin to maintain an adequate link quality (and to maintain demod lock) during the interval between the Eb/No degrading and the ACM controller responding by lowering the ModCod. See the Notes and Recommendations section.
6) Under CONFIG, IP, Switch Set Up, WAN BUFFER, enter the desired size of the WAN Buffer. The minimum size is 20 milliseconds, and is referred to the data rate corresponding to ModCod0. In order to achieve minimum system latency, do not make this value unnecessarily large.
10 | P a g e
June 2009
11 | P a g e
June 2009
The WAN FIFO (the size of which is configurable in the IP Switch setup) produces two control signals that enable and disable the sending of Ethernet Pause Frames. A Pause Frame is an Ethernet frame designed to implement flow control at the MAC layer. A switch supporting 802.3x
Figure 4. The WAN FIFO (the size of which is configurable in the IP Switch setup) produces two control signals that enable and disable the sending of Ethernet Pause Frames. A Pause Frame is an Ethernet frame designed to implement flow control at the MAC layer. A switch supporting 802.3x can send a Pause Frame (with Pause time set to 0xFFFF) to force the link partner to stop sending data. Devices use the Auto-Negotiation protocol to discover the Pause Frame capabilities of the device at the other end of the link. In the diagram it can be seen that when the WAN FIFO reaches a fill state of 87%, it signals the Ethernet Switch to send Pause frames back to the LAN to inhibit the sending of further data. The Pause Frames continue to be sent until the FIFO fill state has reduced to 75%. At this point, normal operation in resumed by sending a Pause Frame with Pause time set to 0x0000. This mechanism has been shown to be very effective at mitigating packet loss when the ACM controller reduces bandwidth.
June 2009
When running in ACM mode the demodulator is performing blind acquisition meaning that it has no a priori knowledge of the modulation type or code rate. For this reason the demodulator acquisition time will be slower than in CCM mode. However, the acquisition time is typically under 1 second for all symbol rates and noise conditions. Running the ACM link with the Target Eb/No Margin set to 0dB will give the best utilization of link power, but in conditions of fast fading may cause demod unlock events, or highly degraded BER just prior to the switch to a lower ModCod. In order to mitigate this, we recommend a Target Eb/No Margin of at least 1dB more if the fading events are particularly severe and/or frequent. The value of Max ModCod may be limited by other FAST codes installed in the CDM-625. For example, suppose the 4100 ksps FAST option is installed, and the symbol rate set to 4100 ksps, the theoretical maximum data rate would be 14 Mps at ModCod 11. However, if CnC is being used, with a 10Mbps FAST limit, the ACM Max ModCod will be limited to ModCod 7, or 9.6 Mbps. Therefore, if for a given symbol rate, it is not possible to set the Maximum ModCod to the desired value, the user should check to determine what other FAST codes may be limiting it. The BER versus Eb/No performance of the ModCods is identical to the VersaFEC CCM modes described in the Operations Manual (FEC Options Chapter). SNR is the preferred metric for driving the adaptation this is the value displayed on the monitor screens. If the user wants to convert this to Eb/No then remember that the relationship is simply Eb/No = SNR - 10log(Spectral Efficiency). To achieve minimum latency, set the WAN buffer to the smallest practical value. The default setting is 20ms, and we recommend keeping it at this level. At this time Comtech EF Data has chosen to disable AUPC while ACM is active. This may change in the future, but for now, ACM should be considered to be a constant power, constant symbol rate scheme. All IP features that are available in the CDM-625 (VLAN, QoS, etc) are available when in IP-ACM mode. The sub-mux feature, however, is not available. VersaFEC ACM is 100% compatible with Carrier-in-Carrier. If required, VersaFEC ACM may be used in conjunction with any of the EDMAC modes, either for serial remote control of the remote modem, or for SNMP proxy. It should be emphasized, however, that unlike AUPC, a framing mode is not required for SNR reporting. ACM maximizes throughput not only when Eb/No varies due to atmospheric conditions, but will also mitigate the effects of other impairments, such as antenna pointing error (due, for example, to operation with an inclined-orbit satellite), excessive phase noise and certain types of interference. However, rapidly fluctuating impairments (~ less than 1 second) such as scintillation at low antenna look-angles at C-band will generally not be improved by ACM. VersaFEC ACM modes are not compatible with VersaFEC CCM modes, due to differences in frame preambles. The CDM-625 was purposely architected to provide the platform for VersaFEC ACM, and has required new approaches to the signal processing employed in both modulator and demodulator. It is the intention of Comtech EF Data to include VersaFEC ACM in future modem platforms.
13 | P a g e
June 2009
System type Interface Remote SNR reporting Max span of data rate Switch point (decreasing SNR) Switch point hysteresis Max fading rate Max ModCod update rate Configurable parameters
Adaptive Coding and Modulation, using BPSK, QPSK, 8-QAM, 16-QAM and VersaFEC short-block LDPC coding. Total of 12 ModCods 10/100 BaseT Ethernet, with auto-negotiated Congestion Control Automatically reported from remote modem built in function at the physical layer requires no additional overhead 7:1 over range of adaptation Corresponds to SNR (Eb/No) that gives BER = 5 x 10-8 0.3 dB Approximately 1 dB/second (higher if Target Eb/No margin > 1 dB) 1 update every 2 seconds (no restriction on distance between ModCods) Minimum and Maximum ModCod (ModCod0 through ModCod11) Remote Demod Unlock Action: Maintain current ModCod or Go to Min ModCod Target Eb/No margin (0 to 4.5 dB, 0.5 dB steps) 54 milliseconds max (for a system operating at 100 ksps, and assuming a WAN buffer of 20 milliseconds, not including satellite path) Tx and Rx ModCods Local and Remote SNR (-3.0 dB to +22.0dB, 0.1dB resolution, +/- 0.5 dB accuracy) Config and monitor menus displaying data rate, modulation and code rate update dynamically with ModCod Spectral efficiency, bps/Hz 0.49 1.07 1.26 1.41 1.61 1.93 2.13 2.34 2.93 3.12 3.32 3.41 Typical Eb/No, for BER = 5 x 10-8 2.4 dB 2.2 dB 2.7 dB 3.4 dB 3.8 dB 4.6 dB 5.2 dB 5.6 dB 6.3 dB 7.0 dB 7.5 dB 8.0 dB Min. Data Rate, ACM mode 18.1 kbps 39.6 kbps 46.7 kbps 52.2 kbps 59.6 kbps 71.5 kbps 78.8 kbps 86.6 kbps 108.5 kbps 115.5 kbps 122.8 kbps 126.2 kbps Max. Data Rate, ACM mode 2.00 Mbps 4.38 Mbps 5.16 Mbps 5.78 Mbps 6.60 Mbps 7.91 Mbps 8.73 Mbps 9.59 Mbps 12.01 Mbps 12.79 Mbps 13.61 Mbps 14.00 Mbps
System latency
Monitored parameters
Modulation BPSK QPSK QPSK QPSK QPSK 8-QAM 8-QAM 8-QAM 16-QAM 16-QAM 16-QAM 16-QAM
Code Rate 0.488 0.533 0.631 0.706 0.803 0.642 0.711 0.780 0.731 0.780 0.829 0.853
14 | P a g e