What comes around ...
"Return Systems 101" (see CED, August 1995, p.66) opened with the theme, "Everything Old is New Again." This article could be paraphrased as "what goes around, comes around." As more networks are being activated with operational two-way signal flow, more questions and ideas surface. Some questions are new and need answers. Others relate to issues and need refinement of past solutions.
While Return Systems 101 focused on basic operation of two-way networks, this article will cover more complex issues relating to setup, troubleshooting, equipment and test systems. The information presented comes from experiences in actual deployments of telecommunications services. Some testing is designed to understand how a problem impacts the network, why it happens and how it is introduced to the network. The great news is return networks work very well in supporting telecommunications. The difficulty lies in understanding how to set a network up for best operation. Information presented here will help with network operation decisions.The network
As previously indicated in Return Systems 101, the return laser is the weakest link in the return path from the subscriber to the receiver in the headend. Guess what? That is still true. The main source of the problem is how the laser responds from desired and undesired signals at its input. The peak level of all signals presented to the laser input must be below the maximum peak power for which the laser is rated. The question is, what is maximum peak level for a given laser? No one really knows. Today, manufacturers do not specify their equipment in this manner. Sometimes, the manufacturer can provide that information, but more often, information is not easily available. Manufacturers are in the process of clarifying how the specifications are written to be more helpful to the operator.
The previous article also recommended using the input to the laser as the network reference level. The idea has proven sound. At the output of the return receiver, it was suggested that the level be set to a convenient point slightly lower than the output from the return receiver on the longest return fiber optic path. All other return receiver outputs would be referenced to the same point. The goal is to preserve as much level at the output of the return receiver as possible to overcome losses in the coupling network behind the receiver.
Recently, with the deployment of cable modems, the thought has changed somewhat. The level at the return receiver output should reflect the input to the laser. For those who have already figured this out, congratulations (sometimes the old brain gets in the way of common sense). The reason centers around knowing how the network is operating. Any signal hitting the input of the laser will be directly related to the established reference. If the network starts to deviate, it will be detected quickly because the level at the output of the return receiver directly corresponds to the return laser input.
That's a nice starting point, but not the complete story. We need to understand where the point of clipping is relative to the lasers used. Currently, most manufacturers provide a specification which does not provide adequate information on how their return lasers operate. This makes aligning the laser and, subsequently, the network, much more difficult. The major concerns relate to the point at which the laser clips and where low inputs become overwhelmed with noise.
Working to deploy fully operational networks, I have seen several instances where high numbers of impulsive hits have been recorded. This led me to take a harder look at what causes problems in networks. What was found was a little surprising. To that end, with the help of CableLabs and Motorola, an operational two-way network was used to test the return of the HFC network. One of the first areas needing attention were return lasers. Below is a discussion of three ways to determine the maximum peak input level to a laser. They differ in the signals used to determine how the laser functions and, interestingly, present non-correlating information. The first attempt at laser correlation was done in conjunction with CableLabs and is discussed below.The laser
In the September 1996 issue of CableLabs' SPECS Technology newsletter, an article on "Testing cable return plant for clipping" appears. The article, by Tom Williams of CableLabs, is clear and will not be related in this article. The premise of the test involves a primary carrier at 8 MHz which is increased in level by 2 dB steps at the input to the return laser until the second (16 MHz) and third (24 MHz) harmonics appear, indicating clipping in the laser. The viewed harmonics could come from single-ended return amplifiers. However, the amplifier input range is generally greater than the laser and not the source of the harmonics. In short, the laser will most likely be the source of the problem.
A simplistic view of this test shows the difference between the manufacturer's rated level specification from the data sheet vs. what level causes clipping at the laser input. That difference is headroom. This test was run in conjunction with CableLabs in a two-way network this past summer. The results proved the tested network operated fine but could be driven into clipping. Further work showed the network was aligned toward the clipping side of the laser. The input to the laser should have been centered between noise on the low level side and clipping on the high level side. Figure 1 presents the idea.
The second method to test a laser involves the use of QPSK data modems and a data test set. The procedure is similar to the one above. Place the modems in the return at a frequency point that allows you to see the second and third harmonic in the return band. Keep increasing the input level until the bit error rate (BER) reaches 1×10-6. The difference between the manufacturer's specification and the level attained in the test suggest the peak input level the laser can tolerate.
Working with that information will allow for the establishment of a power-per-hertz level allocation in the 5 to 40 MHz band. The description presented here is skeletal in nature. For a more in-depth discussion on this topic, read the paper, "Lessons for the interactive return system," presented by Dr. Kerry LaViolette of Philips Broadband Networks in the 1996 NCTA transcripts. Dr. LaViolette's testing differs from the above in that he had data carriers fully loaded in the 5 to 40 MHz return. One aspect of his testing showed that lasers, while performing to comparable curves could vary widely, which further enforces the idea of characterizing all lasers.
The test sequence actually used was designed by Motorola and involves the use of an arbitrary waveform generator (ARB). The designed waveform programmed into the ARB is a tone sequence equally divided on each side of a center frequency and appearing as though it represented the peak power level of a form of QPSK modulation. The testing was run on an active network, and the results proved to be interesting.
First, the standard CW tone test showed the laser under test could operate at a peak input level well above the manufacturer's published specification. However, the engineered test signal from the ARB put the maximum input level to the laser at, or slightly below, the manufacturer's specification. An interesting side result of this testing was directly visible on a spectrum analyzer. Every time a high-level impulsive strike entered the network, it caused CTB to show up around the ARB signal on the spectrum analyzer in the headend. The level of the intermodulation varied with the amplitude of the strike. This proved to be a solid verification of the type of problem prevalent in this specific network. The conclusions taken away from this test are:
- C/N, while needing to be held high, may not be a good predictor of network operation in a data system.
- Return lasers need to be characterized in a number of ways to determine the best median operating point once the peak operating point is established.
- C/CTB may become a more important predictor of network operation than C/N.
- Increasing signal levels into the return on "the more the better" mentality might be self-defeating.
The testing done with Motorola on a specific type of return laser shows this laser can only tolerate six narrow-band frequencies at individual channel levels about +12 dBmV, significantly lower than previously assumed. It needs to be stated that the specifications on this laser are for one video channel with some data carriers running 10 dB below the video. In this case, there are no video channels in this return. The six frequencies combined are smaller than the one video channel. It needs to be stated that other return lasers already perform better, but points up why network laser characterization should be done in every network.
Why is this information important? It sets the stage for a well-operating network. These concepts show how to operate the network as more digital carriers are added to it. As more channels of the same level are added to the return network, the laser begins to see higher cumulative power at its input. As the input power increases, the maximum power input to the laser will be reached or exceeded, which affects the network.
The simple response to this problem is to lower all carrier levels to a point that does not affect the laser. Unfortunately, the simple answer is a little too simple. The digital modulation schemes used for modem, telephone or high-speed transmitters in the return need to deal with peak signals, not root mean square (rms). Digital signals have peak power points much like the peak power point of a television signal. In the case of the television signal, the peak is the sync tip. A 64 QAM digital signal has a peak signal 9 to 10 dB higher than the RMS value. That peak point is represented in a 64 QAM signal at the point farthest away from the origin in the signal constellation.
For a QPSK modulated signal, the peak can be 3 to 5 dB higher than the RMS value. The concern about peak power into the laser might not be a problem with one or two digital return carriers. As more carriers are added, the total power level increases just because more carriers are present. However, the power can increase much higher than the power you see as the carriers are added. In fact, as the modulation schemes come in and out of phase, the peak power can dramatically change. The total power can be pushed to levels well above what the laser can tolerate. This not only raises the specter of clipping and possible laser destruction but can show deficiencies in how the laser handles multiple high-level carriers. The resulting intermodulation problems, which can be caused by desired and undesired signals, will definitely affect any type of communications. The recommendations are:
- Characterize return lasers to understand how they operate.
- Depending how the return looks, run the levels at a point that increases the margin to interference by subtracting out headroom for ingress/impulse problems.
- Determine levels into the laser based on a loaded network bandwidth. As more carriers are added to the network, correcting the whole mess could be difficult as penetration levels climb.
The manufacturers are getting better at helping the industry get the most out of their return networks. An excellent example of this is shown in Figure 8. This chart, reprinted with the permission of C-Cor Electronics, provides the operator with all the information needed to align a return network properly. While it is based on video channel measurements for carrier level and C/N, it provides a great tool for network setup. The chart makes it easy to determine how the network should perform.
In this case, from the bill of materials (BOM) or directly from the design maps, count up the total number of system and line extender style amplifiers. Enter the chart on the proper axis for each to the point where the columns intersect. The C/N listed is what you should expect for the RF plant C/N based on the 17 dBmV flat input levels into the return amplifiers. The reference return optics path loss budget for most manufacturers is in the 5 to 8 dB range. The C/N from the optical path is typically listed as 51+ dB for those budget losses. In a typical 50 0-home node with somewhere between 32 to 64 actives, the RF C/N will be in the range of 48 dB. The combined network C/N will be in the range of 46 + dB, well above the typical low- to mid-thirties C/N often seen in operating systems. Keep in mind that a 46 dB or better C/N number is based on a 4 MHz video bandwidth. The C/N for a much narrower digital signal will be significantly better. Keep in mind C/N may not, by itself, determine total headroom and operating range for the network.
For all intents and purposes, the majority of problems in the return come from the drop systems attached to the hard coaxial networks. This has been verified by various groups working with the return network. I have stated on a number of occasions that 70 percent of the problems come from the subscribers' homes, 25 percent from the tap to the ground block and 5 percent from the hard coax plant. Generally, the 5 percent in the hard coax came from critters, craft and catastrophe. Figures 2 and 3 add another word: corrosion. Both figures are from areas in a system where dynafoam cable is still in the network. Figure 2 shows the superimposed spectrum analyzer image of water in a fitting. This was an intermittent problem, causing the rise in the noise floor by over 30 dB.
Figure 3 shows the response of a wrong type of connector used on the wrong type of cable with a little moisture thrown in for good measure. Figure 4 is what the network looked like after the problem was found and repaired. It does not take a rocket scientist to figure out communications can operate well in Figure 4 and not at all in Figure 3. Return systems can, and do, operate very well. The key to how they operate is practices, procedures and personnel.
The simple response is to lower all carrier levels to a point that doesn't affect the laser.The drop system
I have long held the belief that it is better to understand how something works badly than to understand how it works well. If you understand the "lemon" aspects of operation, you can learn how to make lemonade. With that in mind, one system I worked with proved to be more than a little confounding with a smattering of confusion thrown in for good measure. The network typically operated well, but there were times when trial users reported less-than-optimum performance.
Viewing the nodes in question on a spectrum analyzer, at times they looked fine, and sometimes problems were apparent. The confusion came when the nodes looked fine but the trial users reported degraded operation. A clue to the probable cause came while collecting data on the modems. By varying the length of the packets sent, either 64-byte or 1518-byte, a consistent picture started to emerge.
The 64-byte packets would pass through the network with relative ease and high reliability. The story was quite different for the 1518 byte packets. They would be "hit" frequently and in some cases multiple times. These "impulsive" problem(s) are difficult to find via standard CATV spectrum analyzers because of their sweep speeds. It is highly probable that the problem is not caught because the sweep is at a different point than where the problem takes place. The test equipment needed to find these types of problems must log the occurrence as soon as it happens. Further, if the problem reoccurs, it may leave a signature trace. Realizing the trace has a signature can aid troubleshooting.Detective work
A plan was devised to help narrow down what, how and where these problems were coming from. The plan involved taking baseline information on nodes. The nodes had been balanced, and work was done to clean up ingress sources to a reasonable level. Further, the plan involved purchasing two types of windowed (to pass the return carrier from home convertors) high-pass filters.
One group of filters had an attenuation of 40 dB outside the window, while the other had a 60 dB attenuation. The window is centered at 8.9 MHz for the General Instrument convertors being used. CableLabs was solicited for the test equipment used to log events (hits) as they happened. Four nodes were tested and provided some interesting results. Of the four, one node, far and away, was plagued by random events.
On a per-day basis, the node sustained thousands of hits which could effect BER. Figure 5 is the node prior to filtering. Figure 6 is the node after filtering (so we thought). Note the low frequency end of the spectrum and the drop in the noise floor on the high end. The carrier in the middle of the right-hand side of the screen is used in this testing. The actual data carrier occupies the first gradicule to the left of center screen.
The highest daily rate in a 24-hour period of time was 6,940 events. The estimated BER in the period was in the 10-3 error rate before the network was filtered. The availability of the network, at times, was about 70 percent. No wonder the trial users would see slow operation! Once the network was filtered with the windowed 40 dB high-pass filters, the hit rate dropped to a maximum of 1,572 events in 24 hours — a sizable reduction. But the story does not end here. The BER with the filters only edged into the 10-4 range. The estimated network availability in this node improved to an average 99.4 percent over the 24 hours assuming a threshold of operation of 1×10-5. That number may or may not be reasonable, but it is the point where most forward error correction (FEC) starts to operate.
This node is hardly a sterling endorsement for filtering. I had difficulty accepting those numbers and decided a more thorough investigation was needed.
Discussions with CableLabs and summary information provided the answers. There are, in fact, signatures left by the offending problems. Armed with the new information, a closer evaluation of the node was made. That evaluation led to the following discoveries:
- Short-term, high level CB radio ingress is clipping the laser.
- The RF output from the convertors is clipping the laser.
- Very high level impulsive problems coming through the window are driving the laser into clipping.