JohnSwenson

The 400Mb through put is a built in aspect of the chosen processor. To change it would require a different processor. This is not as easy as it sounds. The specific processor we currently use has two very specific aspects which is what makes this series work. These two attributes are extremely important for the sound of these products. These processor attributes are extremely rare, making it almost impossible to go with some other processor. John S.

Hi tgb, this is NOT about bandwidth of a scope, it is about how serial data streams are sent and received and how measuring devices have evolved to analyze these. It will take me a good part of a day to write this up properly and I don't have a good part of a day free right now. I think I'm going to do this as a white paper and put it on Alex's site, it is WAY to much to put on a forum message. This seems to have become a BIG DEAL so I want to cover this fully and put it in a specific place so I don't have to keep on talking about it over and over again. John S.

Do you have a link to the 10G eye diagram you are talking about? It is most likely the electrical side of an optical link, that IS binary and has a standard eye diagram. You can certainly do the same thing gigabit fiber. The problem is with the copper twisted pair interface. The actual cable pairs have pretty low bandwidth over long distances so they use multiple voltage levels to increase the data rate then use 4 pairs to get it up to gigabit. THAT is the interface that is going to be hard to do an eye diagram for. John S.

I just had an intriguing change in this. My scope had been having some functionality missing recently and I have been working with the manufacturer on it. They finally concluded it was a conflict between two of the the software options installed. By disabling one I could get the other to work. After we got that working they sent me an option key for a brand new serial analysis option that is way better than what was already on the scope. The manual says it supports PAM3 and PAM4 (but not PAM5) so it should be able to work with 100Mb Ethernet! I don't have it running yet, I have to download the software THEN install the key. But maybe just maybe it might be able to do an eye diagram of 100Mb Ethernet. It will be some time before I can get it up and running. John S.

The problem here is that eye diagrams are designed for binary signals, Ethernet is not binary (two voltage levels), 100Mb has 3 voltage levels and Gigabit has 5 voltage. What would an eye diagram for that even look like? The test equipment certainly isn't designed to do it, there is no standard on what a 3 or 5 level eye is supposed to include. Gigabit is about10 times worse because each pair has both directions AT THE SAME TIME on the same wires. This results in receive and transmit (5 voltages) overlaying on top of the other, this can beover 10 voltage layers, and how do you distinguish receive from transmit to include in the eye. 100Mb might be possible to do (separate RX and TX pairs and"only" three voltages to deal with but it is not going to look like any eye diagram you have ever seen. I'm not even sure eye diagram software on my scope can even deal with it. The three voltage levels are going to confuse the PLL clock recovery system. who knows what that is going to produce. And as is the case with every test instrument made it is going to ignore all the low frequency components which is where most of what we are trying to deal with exists. So even if you could actually make one, it probably wouldn't show much of any different, it's not that the ER doesn't do anything but that all the test equipment out there designed to measure such things deliberately completely ignores the aspects we are trying fix. John S.

Floipflops have a very special place in my life. Fresh out of college(many decades ago) I was trying to get job in an Engineering glut. Every open position had hundreds of applicants applying. I had many interviews and I was told that I was something like 200 out of 800 applications, I was getting desperate. I went in for the interview at LSI Logic, they told me what they needed, which was a position that had never existed at any company, something brand new. I told them I was good at learning new things. Then they took me into a room with a big white board and said write down the timing equations for a flipflop. I had no idea what he was talking about, I mentioned that and thought I had just lost the position. He then said "well then derive it on the board". I took the marker and drew the schematic for a flipflop and started writing down all the timing paths, then consolidating them and finally had a simple set of equations. He said "that's it, nobody else has ever been able to do that, your hired!" I was there for 33 years and it all started with a flipflop. John S.

I have covered this topic exhaustively in many posts and some papers on the UpTone website. Digital audio networks seem to be "disturbed" by two different things, jitter on the data and common mode current on the wires. These are actually converted one to the other in many places in the network path. The end result of this is jitter on the clock and data in the DAC subtly changing the analog signal output from the DAC. This has absolutely NOTHING, I repeat NOTHING to do with the data bits being corrupted, check sums etc. It has to do with subtle changes in timing of the data bits and small ground plane differences getting into the DAC through the normal network connections. The Data doesn't get corrupted in any way, the bits are still the same. There are some big misconceptions about optical network connections. The optical connections do NOT make the timing perfect. They do NOT "reclock" the data. whatever jitter that is on the input is still there on the output, in fact the electrical to optical and back to electrical ADDS jitter to the signal. All higher quality optical equipment can do is add less jitter to what is already there, it CANNOT reduce it. What is beneficial is that it completely blocks the common mode noise including leakage currents from power supplies. It blocks one of the issues but makes the other slightly worse. All higher quality optical equipment can do is make that "slightly less" even less, but it is still there. I have a huge amount of detail on the processes involved if you really want to get down into the details. John S.

This thing is really lacking in any kind of technical details, it is hard to figure out what it is doing which is necessary to figure out whether it might be useful. If it is inexpensive or you can get one try, go ahead and try it out. If it is expensive and you can't send it back if it didn't do anything or made it worse, I would skip it. Anything that can improve on a Topaz is very rare, this may be one, but without being able to try it out I would not spend significant money on it. I read a number of their papers and still have no idea what they are actually doing, they like to talk around it without actually saying anything. It may be perfectly legitimate, but I get a funny feeling with their dancing around the details. John S.

I'm not talking about ground plane noise here, I'm talking about noise picked up "through the air" by the cable itself. Even if you had significant leakage current running through the clock cable, a filter wouldn't do any good because leakage current is common mode, it is the same on the ground AND signal conductors, a filter works on the difference between the ground and signal, if they are the same a filter has nothing to work on. The length issue is only applicable IF you have an impedance mismatch and you are running a square wave. With a sine wave with an impedance mismatch the only length issue is being too short. Exactly how short is very much determined by the receiver circuit so there is no way to generalize. My guess is that 2 feet and up is probably fine in most cases, between 1ft and 2ft may or may not be fine, and probably a good idea to not go below 1 ft. If everything is properly impedance matched there is no length issue at all. John S.

Lets tackle the impedance issue first. The problem with reflections occur at the receiver threshold detector, if the reflection comes back to the receiver and is near in time to the original it can cause the detector to give erroneous edges, or jitter to the detected edge. As you mention the sine waves do get reflected as well, but at 10MHz the wave length is long enough that a reflection causing disruption will take either a very long or very short cable. With a square wave the harmonics have shorter wavelengths making it much more probable the reflection may cause a detrimental effect on the receiver. The issue of shielding, filters etc is a bit more complicated. It has to with the "steepness" of the transition, the ramp time. A square wave has a very fast transition, a sine wave a very slow transition. When noise is overlayed on top of the signal, lets say from noise picked up by the cable, the result on the receiver is very different for a slow vs fast transition. Lets use some actual number here to make it easier to understand. Lets say we are using a receiver that has a threshold of 2V, below 2V the receiver interprets that as "low" above 2V it sees a "high". Lets say that there is 10mv of noise overlayed on the signal, the signal is now 2.01V, it will reach the 2.0V threshold slightly earlier, the square wave (with 1ns per volt ramp) sees a 10ps change in where the threshold crossing happens, but with a sinewave of 100ns per volt ramp that same noise causes 1000ps (or 1ns) of change in where the receiver sees the transition. Thus the same noise produces a huge difference in timing change at the receiver. The result is that sine wave signals need hugely better shielding than square wave signals. The filter at the receiver helps to filter out noise that makes it through the shielding. With a squarewave the most important property is wide bandwidth, attenuation of the harmonics decrease the ramptime allowing noise to create greater timing changes in the receiver. Because of fast ramp times extremely good shielding is not as important. For a sinewave, which only has one frequency, wide bandwidth is not important at all, but shielding is extremely important. John S.

Two ways: clean up noise on ground plane and use differential signal. For a differential clock signal you have two wires, when one goes up the other goes down. The "trigger point" is not referenced to ground at all, it is when the voltage on the two wires cross each other. Thus eliminating ground plane issues. Of course many chips only have a single ended clock pin, thus making it impossible to use a differential clock directly. What we do in the ER is use differential signals to send the clock around the board with differential to single ended converters right next to the clock pin of each chip driven by a single ended clock. The is actually almost as good as a true differential clock input. Understanding why this is the case takes a little more info on how this ground plane noise affects a clock. The issue occurs when there is a voltage difference between the ground pin of the source chip and the ground pin of the chip driven by the clock. If the two chips are far apart there is a pretty good chance that all the other stuff happening on the board can produce ground plane noise between the clock source and clock input. BUT if the source is a mm away from the input then there is a much lower chance that there will be significant voltage difference between the source and input. The only noise you will get between them is changing ground current from the differential to single ended converter, and because it has a differential input there is no contribution from other sources, it is just what the converter's output is responsible for. This is not as good as a true differential clock input on the chip, but it is WAY better than a single ended clock running around a board. John S.

Unfortunately, that depends. Independent power domain means power supplies where leakage current cannot flow between the DC sides. Remember that leakage current is low frequency AC so "galvanic isolation" is not sufficient. Anything where the "grounds" of the two supplies are connected is definitely out. Also because of the "high impedance" leakage current (read other posts of mine on that), the impedance between grounds should be at least 1 mega-ohm. What this means is that the grounds should have a capacitance of around 10pf or less. Measuring this is not easy. You also have to make sure there are no "sneak paths", DC power connections from the same supply output going to different devices on opposite sides of the isolation (ER or fiber link). External clock boxes are notorious for this. Two completely separate power supplies meet the requirement. Supplies with multiple outputs may or may not meet this. Unfortunately this is hard to measure. Nobody makes such a tester, so anything has to be DIY and this is not easy. My setup uses some very expensive test equipment and my own circuits. At LEAST the grounds between outputs need to have very high resistance when measured with an ohm meter. This does not mean they are guaranteed to be isolated as far as leakage current is concerned but it is good first step. Hmm, thee might be an inexpensive way to test this, take two outputs, put a 1K resistor on one ground, between that resistor and the other ground connect a battery powered 100Hz sine wave oscillator (NOT a square wave). Use a scope to measure the amplitude across the oscillator, then measure the amplitude across the 1k resistor. If it is 1/1000 the amplitude or less, you are good to go. As an example if the oscillator has a an amplitude of 1V then the amplitude across the resistor should be 1mV or less. Many scopes have probe grounds connected to the AC safety ground which will likely totally mess up this up. Maybe use a battery powered headphone amp. Connect across the oscillator, adjust the volume until you hear a good output, but not blowing your ears off, then connect across the resistor, if you can just barely hear the tone (or not at all) you are good to go. Maybe somebody ought to make one! NOTE!! the positive outputs of the supply are NOT connected to ANYTHING in this test, it is just between grounds. I hope this makes some sense. John S.

Again that depends. A fiber connection is only going to block leakage currents if both sides are completely independent power domains. Fiber connections ADD jitter. Whether that can be taken care of by clocking closest to DAC depends on the implementation, even if there is an OCXO involved, I can't give a blanket statement that it will always be better. John S.

The packet data (sometimes called the "payload") going through your path with the FMCs should stay the same no matter the number of the FMCs or switches in that path. What DOES change is the leakage current and jitter on the data. So If you ignore the leakage current and jitter there is no difference between a single wire and combinations of switches and FMCs etc. The number of devices and cables you add in the path will significantly increase the probability that said payload WILL wind up getting corrupted or stopped all together at some point in time. My recommendation is to only add stuff when it does decrease the leakage current and or data jitter. If you have a very long run (say 100 feet or more) a fiber run in there may make sense. Switching to fiber, or an EtherREGEN in the path may in fact make a significant improvement, but just don't assume it. There is one situation where a switch in such a case may make a huge improvement. This is when different parts of the system are running at different Ethernet rates (100Mb VS 1Gb). This may need what are called "pause frames" to mediate between devices at different rates. Some audiophile devices do not support pause frames but most switches do. This issue mostly shows up when using HQ-Player and an NAA but I think it has shown up a couple times with Roon. This is a pretty rare occurrence. John S.

Your second approach would be the ideal method, just make sure the T-BNC is a 50 ohm model (almost all are). That will properly terminate the 50 ohm clock and cable while a "no load input" is tapping into the signal without changing it. A good way to do this. If it was a really fast signal (GHz range) this would not work, but for a 10MHz signal it is a good method. John S.

This cable is specifically designed for square waves, most of it's "special sauce" is irrelevant for sine waves. Of course you can buy it if you want, but you are spending money on stuff you don't need. For sine wave the best cables you can get are are the "semi-rigid" type. A custom semi rigid cable from Pasternack or other RF companies will be about as good as you can get. Cybershaft also sells such a cable for about a third of the price of this cable. If you do have a square wave that is a different story, then a cable such as this MAY be appropriate. I don't know how good this really is. There are some manufacturers that go with all the fancy stuff, then get the characteristic impedance wrong or use the wrong connector. If you do this all the fancy stuff doesn't do any good. If I was going to spend this kind of money on a clock cable I would want an interdependent measurement of the characteristic impedance of the cable and connector. Particularly if it is supposed to be a 75 ohm cable. You will not believe how many times I have seen 75 ohm cables with 50 ohm connectors! John S.

This is sort of correct, the capacitance has very little to do with a sine wave. The length, sort of. A long cable can pick up more noise which is far more problematic with a sine. So long cables should be very well shielded. So for example the difference in length for a cheap not well shielded cable can actually make a difference. Thus if you want to go a long distance use a very well shielded cable. Even with a square the capacitance per se is not important, it is the capacitance AND the inductance that make up the characteristic impedance that matters. So for a square just focus on the characteristic impedance, if that is correct, the capacitance will be what it needs to be, don't worry about low or high capacitance numbers. John S.

I'm sorry for the confusion, in the original ER the external clock goes through a single ended to differential converter builtin to the clock synthesizer, this WILL convert a sine into a square but not the best in the world. In the ER2.0 the external clock goes into the special very low noise sine to square converter which is a single ended to differential built in. I can't give you exact numbers as to which is better for a square wave since in the ER1.0 it is internal to another chip. Given the performances of the converter circuit and the numbers of single ended to differential converters made by the same company that makes the original clock synthesizer my guess is that the new method will have about the same for sine and square. Compared to the ER1.0 implementation the square will be slightly better and the difference for the sine will be a little bit more. Note none of these are big differences! Maybe along the line of 1-2 dBc at 10Hz offset. Because of the difference in architecture measuring differences between 1.0 and 2.0 can only be done at the output of the clock synthesizer and we know the new clock synthesizer is quite a bit better than the old one, so no matter which clock you use (internal, external sine or square) it is going to be better than what we had in the ER1.0. John S.

The circuit has no way of knowing if the external clock is sine or square. It will work fine if you feed it a square, in that case it behaves as a very low phase noise single ended to differential buffer. So both can be used, the big advantage of this circuit is that a sine will be handled much better [then it is in EtherREGEN “gen1”]. John S.

Comparators on their own are actually terrible at converting a sine to square. Any amplitude noise (AM) in the sine wave gets converted into jitter. The best way to do it is to filter out the noise first then feed it to the comparitor, BUT the comparitor itself generates jitter. The best way is a multistage circuit with low gain at each stage with filters before and inbetween. This is what we are going to do in the EtherrRegen 2.0. John S.

Note that this is only true if the clock is a sine wave. If the output is a square wave the impedance of cables and connectors DOES matter. John S.

Imepedance matters very little for a sine wave clock, but it DOES matter a lot for square wave clock. On running multiple devices from a single clock box, that depends. The external clock on an ER is on the B side, so as long as what you are connecting to the same clock is on the same "domain" as the B side of the ER there problem. The big problem is driving the external clock of the ER and something "upstream" of the ER that MIGHT short the moat. Whether this is is an issue depends on the clock box. If each output is fully isolated (note that galvanic isolation JUST means DC is blocked, the big problem is power supply leakage current which is AC, so a box that says the outputs are galvanic isolated does NOT mean it will block leakage current)

I made several attempts at this but could never build a measurement system that had significantly lower phase and amplitude noise than what I was trying to measure. If your measurement system has more noise than what you are trying to measure no measurements are valid. The measurement system consists of two channels of high bit (24 or more) and high frequency ADCs, being clocked by an extremely low phase noise clock. Just this part is hard enough to do, but not impossible. The hard part is getting this huge amount of data out of the ADC without adding noise to the clock signal and the circuits inside the ADC. I have tried several different approaches that did not succeed. (they get the data out but add too much extra noise in the process) It looks like what I am going to have to do is use ADCs with a very high speed single differential pair carrying all the data, feed that through an SFP+ module -> fiber -> SFP+ module to FPGA with extremely fast SERDES that can store the data to high speed RAM chips, then when sampling is done write SD card etc. Both sides of this need their own battery powered system. The whole thing in a sealed, shielded enclosure. It is getting to the point that all the components of this are coming available without costing half a million dollars. In the next year the parts to do this will probably be available for less than $10K, BUT it will be a long process to develop this. It will happen but it might be several years. John S.

Does the equipment have a clock synthesizer? You need some form of clock synthesizer to get those frequencies from the Mutec. If the device that uses those clocks has built in frequency synthesizers it depends on what the internal clock is and how good the synthesizers are. If the device does not have a synthesizer inside then you need an external one. These can vary all over the place. If you have good synthesizers then a Mutec driving them will probably be better than what you will commonly have inside the device. Without some model numbers there is no way to make any attempt at answering this question. If the driven device does not have clock synthesizers that can run from an eternal reference, then what synthesizers are you looking at? For example the ER includes a clock synthesizer that generates the required frequencies from an external clock, it's built in so an external synthesizer is not required. John S.

I use TrueNAS on a DIY Supermicro server. I love the Backblaze backup builtin. So far no issues whatsoever. It's very important to have a separate drive for OS and TrueNAS with separate backup of that drive. I use NVMe on the motherboard for the OS etc. One thing I do with every computer that is somewhat critical is use ECC memory. Of course you need a processor and motherboard that support it, Supermicro does a superb job with that. Random upsets (mostly from muons produced by cosmic rays) DO happen and ECC is one of the best ways to protect against that.` John S.