December 1, 2005

A description of how new MIBs available from DOCSIS® 2.0 cable modem termination systems can be used to detect and correct impairments on individual cable modems, including group delay, micro-reflections and in-channel tilt/frequency response.

Hal Roberts, Senior Principal Engineer, BigBand Networks; Jeff Finkelstein, Senior DOCSIS Engineer, Cox Communications

Introduction

Cable operators traditionally use RF and packet error MIBs (management information bases) on cable modem termination systems (CMTSs) to troubleshoot and enhance the upstream performance of their DOCSIS infrastructures. Since these MIBs have been available only on a per upstream channel basis, these statistics only provide insight into the average performance of all cable modems. They do not, however, allow isolation of the problems to any individual cable modem.

Additionally, the continuing demand for higher upstream speeds requires the introduction of higher order quadrature amplitude modulation rates , with the result that hard to measure plant impairments, including various sources of linear distortions, often prevent low error rate performance.

The most common method used by cable operators to isolate individual impaired cable modems is utilization of the flap statistic available for each cable modem on the CMTS. Unfortunately, the accuracy of this method is limited because a cable modem will typically only go offline when a total disruption of QPSK maintenance communication between it and a CMTS occurs. Consequently, a flap count of zero can exist even when data communication between a cable modem and CMTS has been completely disrupted .

This paper describes several new upstream RF and packet statistics measured on a per CM (cable modem) basis that allow MSOs to troubleshoot individual modems. These include:

• measurement of the magnitude of CM micro-reflections;
• analysis of equalization coefficients (post-processed for frequency and time domain response);
• signal to noise ratio measurement (per CM);
• CM power level measurements (as seen by the CMTS); and
• error rate calculations using uncorrectable, correctable and unerrored codewords.

It is possible to determine accurately what impairments are affecting any given CM utilizing these MIBs thereby allowing targeted truck rolls and reduced maintenance costs resulting in greater customer satisfaction.

DOCSIS Infrastructure as a Distributed Sensor Array

By utilizing cable modems as signal sources and the CMTS as a multipoint sensor, much can be learned about the nature of the HFC plant impairments. The CM/CMTS system acts as a distributed sensor array, as shown in figure 1 on the following page.

Figure 1: The HFC plant as a Sensor Array

Classes of Impairments

There are a variety of potential impairments in the upstream path, such as:

• additive noise - this affects all cable modems equally;
• linear distortions, such as multi-path tilt and group delay - these can affect each cable modem in a different manner; and
• non-linear distortions - these are rare and are not examined in this paper.

Additive Noise

The most common example of additive noise is AWGN (additive white Gaussian noise). Although this can be caused by many sources it is usually dominated by optical shot noise in the transmitter and receiver in the HFC plant. The SNR (signal-to-noise ratio) is typically a measure of the AWGN versus signal power. Another common type of additive noise is narrowband ingress, and is often called the CIR (carrier-to-interference ratio). Both AWGN and narrowband ingress noise are easy to measure with a spectrum analyzer.

Common path distortion is a non-linear distortion that shows up on the upstream as an additive modulated ingressor at 6MHz spacing. CPD (common path distortion) is measured as CIR. Impulse noise is also an additive form of noise, but as it varies quickly in time it may often doesn't appear in any measurement that averages the signal to noise over a long time period.

All of the above impairments can be adequately represented as per channel measurements, given that all modem bursts arrive at the CMTS at the same level.

Linear Distortions

Linear distortions are caused by poor component return loss, echoes from microreflections, and (differential) group delay caused by diplex filters. These distortions are relatively difficult to measure with a spectrum analyzer and may be present even when channels have very good signal to noise ratios.

These types of linear distortion can be completely mitigated by using equalization, either pre-equalization or post-equalization. In DOCSIS pre-equalization has been specified to avoid the complication and inefficiency of post-equalization. Unfortunately, during the formulation of the DOCSIS 1.1 RFI, which made equalization mandatory, it was not fully understood how hard it would be ensure that all cable modems could be upgraded to benefit from pre-equalization. Even worse, some DOCSIS 1.0 cable modems will not simply ignore the pre-equalization commands but will flap when given pre-equalization coefficients to process. Therefore, although an efficient method of handling linear distortions has been standardized, until all cable modems that are incapable of pre-equalization are removed from HFC channels, pre-equalization cannot be turned on.

In the meantime, the methods described in this paper to characterize and isolate the causes of linear distortions will be essential for increasing upstream bandwidth by making use of higher QAM channels.

Per Cable Modem MIBs

The RF and packet statistics that are discussed in this paper all have MIBs defined in DOCSIS. Unfortunately the MIBs have not been properly defined and as a result have either not been implemented in CMTSs or have been implemented in a confusing and contradictory fashion. The authors have proposed new DOCSIS MIB definitions for microreflections and equalization data that are self-consistent and useful. These new definitions are used in all of the examples reviewed below.

US Signal-to-Noise Ratio (SNR) for each modem.
    - docsIfCmtsCmStatusSignalNoise
US Microreflections for each modem.
    - docsIfCmtsCmStatusMicroreflections
Equalization Coefficients for each modem.
    - docsIfCmtsCmStatusEqualizationData
Upstream Power of each modem as seen by CMTS.
    - docsIfCmtsCmStatusRxPower

All of these MIBs represent information held by the CMTS about specific cable modems. An explanation of what these MIBs mean and how they can be used now follows.

Signal to Noise Ratio

The CMTS does not actually measure the upstream signal to noise ratio but rather the MER (modulation error ratio). SNR is a measure of the power ratio of the signal divided by the plant noise. It does not take into account plant distortions such as microreflections (echoes) or group delay (diplex filters), which can also cause packet errors. MER is a more complete picture of plant impairments as seen by the CMTS. This can be seen in the graphs that follow.

Figure 2: SNR measured using a Spectrum Analyzer

Figure 2 shows a plot of signal to noise ratio measured with a spectrum analyzer. The measurement simply compares the total noise strength with the signal power. Whether the signal is distorted or not does not change this measurement.

Figure 3: MER measured using a Vector Signal Analyzer

MER, on the other hand, is a measurement of all of the significant impairments to low error packet reception, from the point of view of the CMTS receiver. It is based upon the receiver's best reconstruction of the QAM constellation and how far from the ideal constellation the received constellation is. Figure 3 shows an example of this error measurement , with the received QAM16 constellation distorted by the presence of a narrow band ingressor. In this case the MER is 13.9dB.

Figure 4 shows how the error on a given received symbol can be measured.

Figure 4: Estimating Error on a given Vector Signal Analyzer Symbol

SNR for each modem should generally be the same as for the channel. Some reasons why a cable modem may have lower a SNR than the channel include:

• very severe microreflections and/or group delay variation; and
• upstream attenuation too high for cable modem transmit power.

The CMTS commands all cable modems to achieve the same power level at the receiver. However if the upstream attenuation is too high due, for example, to high loss taps then the cable modem will not achieve the proper power level. Although the CMTS is capable of receiving packets that are 6dB or more below the commanded level, the SNR will degrade one decibel for every decibel of power below the commanded power. Therefore, any cable modems with SNR significantly lower than others will usually be experiencing one of these two problems.

Microreflections

Figure 5 shows how two reflecting points can cause an echo that results in transmission symbols arriving on top of one other, known as intersymbol interference. As stated above, equalization can correct for these defects. However, if equalization cannot be used, it is important to be able to measure the magnitude of these microreflections and take corrective action on the plant.

Figure 5: Cable Modem Transmission in Upstream Signal with Echoes

The CMTS is capable of measuring the level of multipath echoes by using a receiver equalizer and processing the equalizer coefficients. These coefficients can be analyzed to show the general magnitude of the echoes with respect to the signal, as is represented in the microreflections MIB, expressed in decibels. The microreflections MIB isolates linear distortions from other random white noise, enabling an operator to tell exactly how severe the distortions are, and whether or not they pose a threat to upstream transmission.

Note that the microreflections MIB is not well defined (as of the writing of this paper) and can lead to improper implementation. This may be corrected in the near-term, and the proposed new MIB definition is used in this paper.

Corrective action can be taken based on the microreflections MIB. The threshold for doing so will change depending on the QAM level of the upstream transmission. Verification of this was achieved by adding multipath in a lab environment and testing the upstream packet error performance versus white noise under different levels of multipath (microreflections). The microreflection MIB levels were monitored as greater levels of multipath were added.

Although it might be expected that the MER (SNR) measurement on a per cable modem basis should indicate when microreflections are causing problems, this is not the case. The MER of long packets will change as the packet is received due to the progressive action of the CMTS receiver equalizer. By the end of the packet the MER may be very good, but damage will already be done in the beginning of the packet. Consequently, MER may remain high when there is damaging distortions.

Figure 6 shows an example of a screen capture from a CMTS, which has severe microreflections causing high packet errors.

Figure 6: Effect of Microreflections on both Microreflection MIB and SNR (MER) MIB

Figure 7 shows how the CMTS receive-side equalizer progressively corrects distortion on a long burst. The beginning of the constellation is distorted by intersymbol interference, though by the end of the constellation the equalizer has converged and the constellation is free of distortion, and has a good MER. However, if the beginning of the burst is distorted, uncorrectable errors may result. The CMTS MER/SNR measurements average the MER over the entire burst, and will not respond to the damaging effect of the distortions.

Figure 7: Improvement in SNR/MER on Long Bursts due to Receiver Equalizer Microreflections - Recommended Allowable Levels

The effect of microreflections on QAM16 packet error rate performance is shown in figure 8. These curves can be compared to the AWGN only curve where no microreflections are introduced. When multipath is increased from 18dB to 9dB the system's ability to withstand AWGN is reduced. However, until the multipath MIB readings reach 15dB, there is very little effect on performance, because the equalizer in the receiver is capable of correcting minor levels of linear distortions.

Figure 8: Effect of microreflections on QAM16 Packet Error curves

Figure 9 shows the same performance with respect to a QAM64 burst profile. In this case the microreflection threshold where a significant shift in the curve is seen, occurs at 20dB.

Figure 9: Effect of Microreflections on QAM64 Packet Error curves

Received Power Level During normal operation of CMTS/CM systems, the CMTS commands a CM to transmit at a pre-set desired receive power at the CMTS. Performance may suffer if any CM is unable to adjust to the power level commanded by the CMTS. Although transmit power level MIBs from the cable modems themselves are useful, it is only possible to determine if a CM is unable to reach the desired receive power level at the CMTS by measuring the CM bursts from the CMTS reciever. For example, by utilizing MIBs from the CMs themselves, a histogram of the CM transmit power levels can be obtained.

Figure 10: Typical CM Tx Power Level Distribution

It can be seen that there is a large drop-off in the number of cable modems with transmit power greater than 54dB. This is a strong indication that many modems cannot reach the desired level of power at the CMTS. However if we wish to know just how far off the modem is from ideal Rx level, we need to use CMTS measurements available from the docsIfCmtsCmStatusRxPower MIB.

Received Power Level - Recommended Allowable Offsets

A print out of the receiver power level MIB is shown in CLI format in figure 11, for two modems on the same channel with two different received power levels. In this case the receiver of the CMTS is set to 0dBmV. One modem is at that level while the second modem is at -6dBmV. Therefore the attenuation in the upstream is 6dB too high for that particular CM. The consequence of this excess attenuation is that the SNR for that modem will be worse. In this case the SNR is 5dB lower than the modem with adequate transmit power.

Figure 11: CMTS Received Power Level MIBs

Ideally no modems should be operating more than one decibel from the CMTS Rx power setting. In practice operators may wish to take action if the power level exceeds 2dB from the ideal set point. However, it is much more likely to have modems operating below the CMTS set level due to the CM transmitter limits and upstream high attenuation rather than the reverse situation with the modem unable to reduce the power to the CMTS set level.

The operator has the following options for correcting this problem:

• eliminate excess attenuation in the subscriber home by locating the CM near the entry point into the home or the ground block;
• reduce tap value at the tap where the drop is connected;
• reduce CMTS receive power setting; and
• reduce attenuation at the optical node.

The last two sets of MIBs that will be examined are those of the equalization coefficients and the packet error statistics.

Equalizer Coefficients and Frequency Response

The multipath MIB is a value calculated from the equalizer coefficients in the CMTS. This MIB reduces the complexity of 8 to 24 pairs of complex tap values to a single estimate of the power of the distortion to that of the main signal. However, more information can be extracted from the coefficients, enabling an accurate picture of the frequency response of the channel to be obtained This will help isolate the source of the linear distortion, typically either multipath or differential group delay.

Figure 12: Normal Equalizer Coefficients

This printout shows that the microreflections are very low, 38dB. Post processing of these coefficients enables the following graph in figure 13 showing the channel frequency response and the magnitude of the equalizer coefficients to be obtained. In this case the channel ripple is very low, only 0.25dB. No maintenance is necessary for this plant. Note that only one coefficient is at a high level (0dB) that of the main tap; all other coefficients are 40dB or lower.

Figure 13: Frequency response of a Channel with Low Multipath

Alternatively, the example in Figure 14 shows high multipath levels of 12dB.

Figure 14: High Multipath Equalizer Coefficients

Note the very high levels of side tap power, especially the fifth tap.

Figure 15: High multipath levels of 12dB result in a 10dB ripple in Frequency Response

Packet Error Statistics

Finally, the ultimate measure of performance is codeword error rate (CER). Although not very useful in tracing the source of the plant problems, CER alerts the operator to the fact that there are problems. In the case of single modem CER statistics the modem experiencing poor performance may also be pinpointed. By extracting the codeword statistics from Figure 12, four correctable errors, one uncorrectable error and 44 codewords with no errors can be observed.

The definitions and MIB names for these are shown below:

MIB name: NonErr - docsIfCmtsCmStatusUnerroreds
MIB definition: all codewords that had no errors from the the modem whose MAC address was used in the 'show modem' command. Note that there may be multiple codewords per packet so these do not correspond directly with Packet Errors.

MIB name: CorrErr - docsIfCmtsCmStatusCorrecteds
MIB definition: these codewords have errors but are corrected by the Reed Solomon decoder. They can be considered an early warning signal for uncorrectable errors.

MIB name: UnCorr - docsIfCmtsCmStatusUncorrectables
MIB definition: these codewords have too many errors to correct.
You may calculate Codeword Error Rate (CER) by the following formula:

The following formula can be used to arrive at the codeword error rate in percent:

CER (%) = 100*(UnCorr/(NonErr+CorrErr+UnCorr))

For example, in this case the CER = 100*(1/(44+4+1)) = 2%

Summary and Recommendations

The following are nominal recommendations for MIB thresholds at which the cable operator should initiate some kind of plant maintenance. These are approximate levels and may vary depending on the customer expectations of the service level. For example, the cable operator may wish to initiate action when codeword error rates are less than or more than 1%.

Per CM SNR

When there is significant CM to CM variation e.g. +/-2dB for SNR, especially when SNR < 25dB the MSO should initiate plant maintenance. What kind of maintenance will depend on what the following MIBs indicate.

Microreflections

Maintenance should be performed if:

• MR worse than 15dB for QAM16
• MR worse than 20dB for QAM64

The type of maintenance will depend on whether the linear distortions are caused by microreflections or by high group delay. Post processing of the equalization coefficients (see below) will indicate which is the source of the distortion. Alternatively, it can be assumed that if the upstream channel frequency is below 36 MHz, then group delay should not be a problem.

Codeword Error Ratio (CER)

Error ratios higher than 1% should trigger CM maintenance.

US Power Level (from CMTS view)

The CM Power Level should not be more than +/-2dB from CMTS receive level (typically set to 0dBmV) If the receive level is off by more than 2dB, it will likely be due to a cable modem at maximum power. The corrective action will usually be to lower the tap attenuation or eliminate in-home splitters between the CM and the drop cable.

Equalization Coefficients

If the above Microreflections thresholds are exceeded, post-processing of coefficients can help direct maintenance to the right solution. Microreflections may mean tightening connectors, group delay may require replacement of diplex filters. Frequency response curve will distinguish between group delay and microreflections.

Click here to download PDF