Updated November 3, 2026: Revised for SC-QAM, OFDM, and OFDMA. Historical comments remain below; some refer to the earlier explanation.
You open a modem status page or CMTS dashboard and see thousands, perhaps millions, of corrected codewords. How bad is that? Before I would call it a plant problem, I want to know what the counter measures and how long it has been running. A big number alone does not tell us whether service is getting worse. Letโs walk through what these counters mean, how to calculate them correctly, and what to check next.
Codeword statistics help locate impairment when treated as receiver measurements over a defined time window. They do not replace MER, spectrum, profile, traffic, or customer-impact data.
What codeword counters measure
A DOCSIS codeword is a block of encoded symbols carried through the physical layer. Forward error correction adds redundancy so a receiver can recover some corrupted symbols without asking for a retransmission. In legacy single-carrier QAM (SC-QAM), ReedโSolomon (RS) coding is part of that process. The receiver counts blocks that arrived without errors, blocks it corrected, and blocks it could not correctโwhen the vendor exposes those categories with those definitions.
โReceiverโ depends on direction. For downstream, the modem reports what it receives; for upstream, the CMTS reports what it decodes from modem transmissions. A CMTS upstream display is not a modem downstream measurement. Identify interface, direction, and reporting object before comparing values.
RS corrects symbol errors, not an arbitrary count of individual bits. With r parity symbols, a conventional RS code can correct up to floor(r/2) unknown symbol errors in one codeword. A symbol is commonly one byte in this context. In a toy example with 16 data symbols and 2 parity symbols, the 18-symbol block can correct one erroneous symbol. Eight flipped bits in that same symbol still represent one symbol error; one flipped bit in each of two different symbols represents two symbol errors and exceeds that exampleโs capacity. This 18-symbol example illustrates the rule; it is not a claim about a DOCSIS-mandated block size.
FEC does not suddenly switch on when a customer packet exceeds 16 bytes. It operates on the defined encoded blocks. The correction limit depends on the code parameters, and failures beyond that capability can leave corrupted data unrecovered.
Read counters by technology
A counter named โcorrectedโ does not mean the same thing on every DOCSIS channel. SC-QAM and DOCSIS 3.1 OFDM/OFDMA use different FEC chains and management semantics, so equal-looking percentages may represent different decode stages.
| Channel and direction | Receiver | What to check |
|---|---|---|
| SC-QAM downstream | Cable modem | Legacy RS codeword counters reported by the modem; confirm the vendorโs exact object definitions. |
| SC-QAM upstream | CMTS | Legacy upstream receiver counters. Some โunerroredsโ objects count only error-free codewords, so corrected and uncorrectable counts must be added for a total. |
| OFDM/OFDMA | Modem downstream / CMTS upstream | Downstream OFDM data uses concatenated BCHโLDPC; upstream OFDMA uses LDPC. Check the direction, profile, and MIB definition. Total-codeword counters are normally inclusive; do not add corrected counts again. |
For downstream OFDM data, CableLabsโ PNM primer, section 5.4.5, describes concatenated BCHโLDPC FEC. In that example, codewords that fail a pre-LDPC syndrome check but pass BCH can count as corrected. Depending on the counter definition, corrected percentages near 100% can occur normally. This is not directly comparable to an SC-QAM downstream post-trellis RS rate. Consult the MIB and device documentation for the specific object and measurement point.
Upstream OFDMA uses LDPC without that outer BCH stage. CableLabs describes corrected codewords as failing the pre-decoding syndrome check but passing the post-decoding check, and unreliable codewords as failing the post-decoding check. An unreliable codeword can still contain error-free data bits, so use MAC CRC results and service measurements to establish the impact. This is another reason to read the field definition before treating every dashboard label as interchangeable.
For SC-QAM, Ciscoโs return-path monitoring guide and the MIB provide useful context. One example is docsIf3CmtsCmUsStatusUnerroreds, which counts codewords received without error. If that is your โgoodโ count, it excludes corrected and uncorrectable codewords. Some OFDM total-codeword counters, by contrast, are already inclusive. Do not add corrected and uncorrectable counts to a total unless the definition requires it; doing so double-counts codewords.
Calculate an interval rate
First confirm whether you have interval counts or cumulative counters. If you can obtain deltas over the same interval for error-free (E), corrected (C), and uncorrectable (U) codewords, and those three classes are mutually exclusive and exhaustive, calculate:
Uncorrectable rate (%) = 100 ร U / (E + C + U)
The corrected rate under the same definitions is 100 ร C / (E + C + U). If the vendorโs โtotalโ already includes all codewords, use that total once as the denominator. Do not add the categories again. If the definitions are unclear or overlap, stop before publishing a percentage and resolve the counter semantics first.
For a worked interval, suppose a channel reports 1,000,000 error-free codewords, 20,000 corrected codewords, and 500 uncorrectable codewords. The total for this calculation is 1,020,500. The uncorrectable rate is 100 ร 500 / 1,020,500, or about 0.049%; the corrected rate is about 1.96%. These are rates of codewords in that sample interval. They are not packet-loss or bit-error rates.
The original articleโs CMTS sample showed 13,861,878 good, 760,804 corrected, and 47,090 uncorrectable codewords. If these are mutually exclusive categories, the combined count is 14,669,772. The uncorrectable share is about 0.321%, and corrected is about 5.186%. Without aligned start and end times, these historical figures do not establish a current rate or service condition.
For cumulative counters, subtract two aligned readings. Check uptime, interface resets, profile changes, and the discontinuity indicator; SNMP counter discontinuities and wrap behavior are defined in RFC 2578. Counters may wrap at their defined maximum or reset; confirm the event and handle wrap using the counter width before calculating. If a reset or discontinuity occurred, discard that interval and establish a new baseline. If no codewords were counted, the denominator is zero: report โno samples,โ not zero percent.
What the number can and cannot tell you
An uncorrectable codeword means the receiver could not recover that codeword at the measured FEC stage. It does not, by itself, tell you how many IP packets were lost. A packet can span multiple physical-layer codewords, and counters may aggregate traffic for many modems or channels. Higher layers may recover some impact through TCP retransmission, QUIC recovery, or application retries; latency-sensitive traffic may experience delay, jitter, or loss differently. Correlate the physical-layer interval with packet, transport, and application observations before describing customer impact.
I treat a 1% uncorrectable codeword error rate as a practical action threshold. At and above that level, users can begin noticing service degradation, particularly with real-time applications such as voice. This is a well-established troubleshooting threshold, not a universal boundary where every service suddenly fails. Ciscoโs legacy upstream FEC guidance flags 1% and above in red and warns that voice quality can be affected at that level. Do not wait for 1% before investigating: burst errors and sensitive applications can cause noticeable problems below it. Calculate the percentage over a defined interval, confirm the counterโs meaning, and correlate it with packet loss, latency, and the customerโs symptoms. This threshold concerns uncorrectable codewords, not corrected codewords; do not apply it blindly to OFDMA unreliable-codeword counters.
Likewise, a cumulative corrected counter that rises every day is expected on an active channel. A larger absolute number does not show that conditions are worsening unless you compare normalized rates over comparable intervals and account for changes in traffic volume, channel configuration, uptime, and measurement coverage. Trend interval deltas, not lifetime totals.
Troubleshoot with correlated data
When a rate changes or uncorrectables appear, shorten the collection interval enough to localize the event. Compare aligned codeword deltas with receive MER, spectrum captures, ingress or impulsive noise indicators, clipping, reflections or micro-reflections, and channel/profile changes. For OFDM, examine RxMER by subcarrier and the profile assignment over the same window; a channel-wide average can hide a localized impairment or a profile transition.
Check whether affected channels, service groups, or neighboring receivers show the same time pattern. A shared event points toward a common plant or node condition; a single receiver may indicate a local drop, connector, or device issue. That is a lead to investigate, not proof of root cause. Verify with field measurements and traffic symptoms, and preserve the time window and device state so the evidence can be reproduced.
I recommend starting with a short, repeatable interval, documented counter definitions, uptime, channel/profile identity, and synchronized MER or spectrum data. Then you can compare before and after a plant action.
Frequently asked questions
Are corrected codewords a problem?
Not automatically. FEC is designed to correct errors, and some corrected-codeword countersโespecially certain DOCSIS 3.1 OFDM countersโcan be high during normal operation. Confirm the decoding stage and trend the rate over a defined interval alongside MER and other evidence.
Do uncorrectable codewords prove subscribers are losing packets?
No. They show that a physical-layer codeword failed at the measured receiver and FEC stage. They do not map one-to-one to IP packets or prove a user-visible loss event. Correlate with packet or application data and the affected traffic path.
Can I compare an OFDM corrected percentage with an SC-QAM percentage?
Generally, not as if they were the same metric. The FEC chain and counter definitions differ, and a DOCSIS 3.1 corrected counter may represent a different stage than legacy RS. Compare like-for-like channels and documented objects, or report the raw values with their definitions.
For deeper context, see OFDM and SC-QAM codeword calculations, DOCSIS timeout descriptions, and T3 timeouts and service impact. For help correlating telemetry with plant conditions, contact Volpe Firm about broadband consulting.
Brady,
Informative as ever-I look forward to the next article on codewords. One thing I’ve never fully understood is the relationship between codewords and packets.If we say the “average” packet is 1500 bytes long, in general, how many codewords would need to be generated to get this packet from the susbcriber, to the CMTS? Other articles suggest codewords being about 256 bytes “worth”, but you mention 16 bytes-is this the same for every codeword? And then we can ask how the likes of concatenation and fragemention affect these as well.
For me, this really is the nuts and bolts. and would then let me understand fully the realtionship between packets and codewords, and then (maybe) come up with BER figures, as well as the current MER that we use.
Cheers
Hi Sandy,
Thanks for the feedback. I’ll hold off on the math for now as its more complicated than what easily fits into a comment reply, but as you suggested, this would make for great article, so I’ll put it on my map.
-Brady
Hey “Sandy”,
It took me awhile to figure out your code name until we shipped you the posters. But now I know. ๐ Hope those make it to you okay.
Thanks for the feedback on the article. I used 16-bytes in this article because I wanted to be able to visually explain the matrix. Using larger packet sizes made it hard to fit on the page. Give me some time and I’ll do a follow-up article to describe in more detail how this scales. I was planning to do it in MATLAB and provide the code so that folks could download the student version if they did not have a copy and play with it.
However as a quick answer, you can take the math I did in the article and scale it accordingly for the first equation under “How Many Errors can FEC Correct?”. Also, there is a relationship between this, codeword errors BER. I’ll include that in a future article also. Sorry about being slow on articles lately, but my clients have been keeping me busy.
Regards,
-Brady
Brady,
Got the charts, many thanks! Look great.As for the Sandy part ๐ , its my middle name, which I prefer to use on public sites.
Thanks for the reply on the codewords, I’ll await a further response when you get the time..
It’s a hot topic over here at the moment, and all the docs I’ve read have never been able to slice up a Ethernet sized frame into codewords, thereby giving us some idea how many codewords “lost” translate into actual packet payload-and therfore the customer experience.
Cheers
Sir,
What is the best method of tracking FEC in a Cable System? Currently I am Pulling the Fiber to determine with Node the Issue is coming from(because we are 4 to 1). Most of the time when I pull the fiber the FEC issue goes away, but I would like to know if there’s away to track it without pulling the fiber in the Hub so that I can actually fix the issue causing the FEC. Any suggestions? Also once I determine the Node I am pulling return pads to track it from there.
Thanks,
Hi Scotty,
When you say “What is the best method of tracking FEC…”, I am assuming that you already know that you have correctable and uncorrectable codeword errors because you are in the headend and pulling on fibers. You are likely seeing the errors by looking at an SNMP monitoring system or by looking at the SNMP CLI, correct?
I generally recommend using a test instrument such as a spectrum analyzer or return path monitoring system at the same time that you making any type of changes in the headend (i.e. pulling on return path fibers) to see what impact that is having on the return path spectrum. In your particular case, it sounds like you could be dealing with dirty, loose or damaged return path fiber connectors. I wrote an article that covers some of this. You can check it out HERE.
A good spectrum analyzer that allows you to display multiple traces (peak hold, live trace, average, and minimum hold) like the AT2500 from Sunrise Telecom is very helpful for any type of live plant testing. You need something very responsive.
Pulling return pads is also a pretty common troubleshooting practice. It can be disruptive to the subscriber, especially if you have Voice subscribers. Plus you likely have to do this during a maintenance window, which means yet another night of no sleep for you. I recommend one of two methods for this which is better for you and the subscriber. Several vendors offer equipment that let you inject a return signal to a headend monitoring system which will check for error rate plus impairments. These work like DOCSIS signals (one is integrated into the DOCSIS network itself). All have their pros and cons, but its a lot better and provides more information than pulling pads.
-Brady
Hello Brady –
Relative to computing the percentage of corrected and uncorrectable codewords, I’m questioning the math in your example:
Codewords: 13861878 good 760804 corrected 47090 uncorrectable
You used the ‘good’ codewords as the total codewords. However, if corrected and uncorrectable are stated separately from the good codewords, would not the total codewords be the sum of the three?
Thereby, the % of uncorrectable codewords would be (47090/14669772)*100 = .3210%
rather than (47090/13861878)*100 = .3397%
In this example, the difference is very nominal; however, for some, this difference may end up being quite significant.
Fantastic job on your blog! Your time and effort is most appreciated!
Hi Elaine,
This is a very good point you make and I glad that you brought it. I agree with you that in some cases it will have a nominal impact. However if you reset your counters on a daily basis and have high codeword error rates, then the impact could be greater. I am now interested to find out if anyone is using an algorithm other than the one I mentioned in my post that takes into consideration what Elaine mentioned.
I will look into this, but if anyone reading this is away that their correctable and uncorrectable error dashboard removes the errors from total codewords first, before dividing please post back here to let us know.
-Brady
Actually, in our product. we use CER = uncorrectable/(good+correctable+uncorrectable). we are not remove errors from total codewords.
Brady,
I think there would be merit in explaining how USCER data is different for upstream and downstream. The systems we use to monitor give us the uncorrectable rates, usually in scientific ratios, 5.21E-7 for example. This is based on the math you explained above. The upstream isn’t a constant flow of data, and after correcting a fault causing errors, a quiet modem will not report an improved error rate until it passes enough data (thru station maintenance) to make an accurate calculation, yes? The same holds true for that modem’s reported USSNR, right?
Obviously, the downstream is different because of the constant flow of data to every device is used for these calculations, whereas the upstream is bursty, and as I put it, “modems only talk when spoken to”.
Hello Brady
When considering metrics via SNMP would you suggest capturing each instance of a given metric or just capture the one instance and use that for the overall measurement? Performing statistical analysis on each instance will be a very thorough approach but I’m not sure that the added complexity is really necessary. Your thoughts?
Hi John,
you can never take a specific measurement in time and assume accuracy. There are too many variables. Itโs best to have a start point and an end point to do delta readings in a specific time period.
Hello Brady
Could you recommend acceptable Upstream Codeword error ratio, correctable ratio, uncorrectable ratio?
1. per cable modem
2. per upstream channel
Thanks
Jerry
Hi Jerry,
Thanks for reading and the question. Your question was selected to be part of our podcast. You can watch or listen here. https://volpefirm.com/docsis-cmts-best-practices-recommendations/
Thanks or your interest.
Mia
Hi Brady,
We had a internal discussion about CER and CCER and stumbled upon this topic on your site.
Thanks for all the effort you are putting into the socials
your formule is:
correctable codewords = corrected codewords/total codewords*100
The calculation on this website is
correctable codewords = 760804 /13861878 *100
In our opinion this should be
correctable codewords = 760804 /(13861878+760804+47090 ) *100
As the total is 13861878 good 760804 corrected 47090 uncorrectable = 14669772 in total
Or do we miss the point somewhere.
Regrads
Bram and Ronald