If where I am here with bar from Clevis Co you've seen before. Billy We've got a new we have a new product. Yeah, we just a new Clintus Coke. Tell us about it.

Okay, it's It's very big and chunky compared to our old one which is sitting over there. That's the original one. Yes, Yeah, Well, that's it. Yeah, the gold edge.

He has four channels. It's 500 mega samples at 14 bits at 14 bit 14 bits and it's New Zealand Bird Spits Spits. It's an isolated channel of oscilloscope. This is what I'm excited to eat.

Channel Each channel has got a kilovolt isolation between itself, the other channels and ground, plus or minus the kilovolt motor drive. We are probing the innards of a motor drive and we're looking at the gate drive and also at the saturation voltage of a transistor that's inside the gate drive, which is going up and down three hundred and twenty volts. There, It's all rather noisy. Here is the voltage that's driving the gate and threw a little resistor which goes to the gate itself, which is that signal there.

So we're looking across a resistor which is going up and down three hundred and twenty volts and a few nanoseconds because you can measure across the resistor because you're isolated and floating. Where there's two requirements here. The first is that is isolated and floating and the second one. the very important one is that it has very high common mode rejection ratio.

Yes, what are you talking about? Throw some figures at it. Well, the CMRR which is how much noise gets into the measurement from the rapidly slowing signal. It's a hundred DB at fifty megahertz at fifty Meek's so it will carry on pretty linearly after that. Unfortunately, my little test rig only chicks up to 50 minutes.

So what's the actual analog to 200 megahertz bandwidth? There's a graph of CMRR over frequency. We got a 65 megahertz and this the hundred. So we're actually a boy or a hundred. beautiful which is not bad and we can.

We can see the effect, the effect of their over. Here we have a full bridge. There's a gonna sleep around a lot. There's a full bridge and what we're doing our pointed things.

We're looking at this gate here which is connected to a switching node which is switching between zero which is the bus and five hundred volts and five hundred and over. Here is another gate which is doing the opposite between zero and five hundred and we've also got two probes looking at the actual switching more than just let me see what they are. Do you Not do that with the regular scope people? I've done a video on how not to blow up your oscilloscope. That's a way to blow up your oscilloscope.

That is a very good way to live. A lot of stuff except if you've got an isolated scope. Our isolated Channel scope like this, Exactly. Yeah.

So here we have a grant. If you can see the connections heated, the two red probes are the switch voltages which are going 500 volts and the two grey probes are connected to the gates. and here we can see they the red one is the 500 volts and switching. So wait, that is very fast.

Yeah, then these signals. Here are those two dates on top of this. Yep and we can see some where you can and win a transition. we can go and investigate.

Ah, Okay so here. Also we zoomed in. This is a zoomed in here. This is a zoomed in view.

So this is looking at that spot there. Yep, so we're looking at that. looks like nice. Well that's that.

It's not. But it's not. No, no, it's the gate turning on. This is maybe I like this, the Mula plateau and that's the the voltage level at which the hold stops rising because the charge is going into the gate like crazy.

Okay now there is a problem with Fitz which is what this is. These are fits in that there's a capacitance between here and another one between here and here. and that capacitance acts like a voltage divider. and as you raise and the effect of it is that these slight expressions from the gate down to the negative bus as you're raising the voltage and the opposite way run when you're reducing the voltage and that tends to suck charge out of the gate drive.

So this reduction here is a reduction because of the Miller capacitance. and this pulse here is because of Mill expectancy is injecting charge back into the gate. These can be problematic if your design is no good because if you make this too low. in fact the low 4 volts, it will turn the Fiat off of course.

Yes! and if you go, it's not good. Same here. If this goes above 4 volts, you start getting a little bit of turn on in the face, then you have to Fitz on at the same time. Oh, is that a good plan? You do.

They do happily. Fitz Actually pretty rugged and as they get hotter their resistance increases and like IGBTs and so you are a little safer. but it's not a good thing to do. the losses go up.

Don't do it. But the big thing is is that this is almost impossible to measure using any other method. You could buy a Tektronix Tivan probe to $18,000 optical isolator and you could use one of these which does the same thing. Yep, except a bit noisier.

At the same, there's 200 megahertz as their lowest cost one right? Only $30,000 and there. and they have a gigahertz one as well which is only $23,000 and it's just a pro. It's just one channel. It's just one job.

That's right, we have four. Yep, okay, no other way of measuring. This does come with five voltage. It comes with every probe you need.

So we give you four. low-voltage one switched one by team by probes and for high voltage 100 times prize. And if you really want, we can give you 200 honest probes because then you can look at 1.6 kv. Excellent yes, but most mosfet solutions stop it around about 5 600 volts? Something like that, that's right, but there are.

There's a lot more technology coming along. They have things called sick. Have you heard silicon carbide bits? Okay, and they're rated at 1200 volts and you know people are stunned to use their 1200 volts. I Don't think there are around when I was doing 500? No, not at all.

They're new then you. and and they are. They are tricky because you can get edge rates down to about 3 nanoseconds Horrendously fast. Well then you have to deal with lots of other problems of what as well.

Of course you know any series inductance, massive ringing voltages you don't want that in the EMC issues you might have. Well, that can be magnified exactly Even even this is a bit fast. I Mean you know seven to eight milliseconds or whatever it was? yeah, was seem. you know I should say it's it's pretty damn fast so that's impressive.

So how much does this cost? Well, this one there the Wii is nine thousand, Six hundred bucks. Yep, moderately expensive Yankee bucks. Yeah, pretty expensive, but we are open to deal. It's a specialized bit of kit and as I say, you can buy a whole one for less than the price of one tech probe.

Now is there any other competition for an equivalent for channel isolated scope? No on the market? No, No, there's no other competition. The T The TPS serious copes. They're isolated, but there are only 300 volts isolation between channels and the common mode rejection is about 30 DB at 10 megahertz. So 33 is like of 51 to 50.

So if you're looking at five hundred volts, you'd be 10 volts of common-mode noisily working around here, you wouldn't see that signal with ten volts of common mode noise. So that's the big deal. That's a big deal. Ten volts or noise? You? You couldn't see anything at all? No, no.

and the same goes for differential probes. Differential probes are also used to try and do this job, but they also have terrible common mode rejection ratio at high frequencies. They're great at 50 Hertz but once you get them up to ten megahertz, they're useless. This 8 nanosecond, these eight Miss Menace' can slew rates.

Well, they're equivalent. Bend with is one of a PI TR So that's one of our about 3 times 8. that's 1 over 24 which is about 40 megahertz about. So that's equivalent of a 40 megahertz signal.

So you have to have pretty good common mode rejection to reject it and not see it in the signal. Yeah, Yes, yes, we can I can. All right. Oh no, no, we.

we do everything. But we're going to turn off because this. There's lethal voltages in here. Okay, so we're unplugging things.

Okay, this right? Sorry I Just want to see the isolation. It's down. Oh yeah, yeah, they're actually yeah. I can see the slots.

Look at that. each one is and Matthew would you say if 1500 volts isolation? Yeah, it's 10 millimeters. Which under the IEC standard, it's about 8,000 volts of real isolation. But in terms of meeting the standard gives you a thousand volts and I can't agree.

3. Yep, Yep, so it's it's pretty good. Sorry, we're not gonna be out right in the cancer. No, we are not.

Oh no, that's that's quite a drama opening the cans. Okay, I'll tell you all about it. So here's these are the four channels. They're isolated from each other.

As you can see, there's a fiber optic link here. These are -. These These. There's actually four fibers in there and they go at ten.

the cable of King gigabits per second. but we only run them at five, right? They're five gigabits per second from here to here and this two channels one way and two channels back the other. These things here. see those little wars.

Yep, there we go. I Couldn't say it right. The color was a bit different. Yeah, yeah, that's the fiber channels.

You'll find that the blue ones go one way and the green ones go the other. And they're running at five gigabits per second. So you've got the convertor inside the front end. Yes, we do.

Just running the digital. I Was gonna ask how are you isolated digitally or analog digitally because we want to minimize noise to the absolute maximum. That is the only choice. That's the only choice.

The other guys, that's the Tektronix guys. They do it the other way. They modulate us a signal and send it as light as a modulated analog signal. But the result of that is that they have much more channel noise than we do.

So we we wanted to keep the channel as quiet as possible and the only way to do there just have as few bits in the way as possible. Yep, that's what we do. So 14 but 500 mega sample ADC plus some and then this. This is the power supply that runs it now.

Power supplies are actually really tricky. We have rolled our own. Now there's the big one. The big issue with us and I'm giving away trade secrets here.

It's like it's so much better. That everyone knows is that the this power supply here has very low common mode noise going out into the real world. it's less than 100 microvolts. nice so it doesn't need bootstrap capacitors to link the input to the output to route the the noise back to the input.

And it's It's driven with a very symmetrical power switch. It's it was hard work. It took us a year to design Just the person Ludie Terrible. Okay, and then as we carry on we you'll notice that we've got two fibers one way and to the other.

Well, one of those as we use for a clock. A very accurate clock. It's a half 1/2 a ppm clock. Yep, it's only running at 500 kilohertz and it comes back.

It comes into each device and it's distributed using a you can't see it, it's on the backside. It's used distributed with a very low latency clock. There's a lot of synchronization issues across the channels. That's another really hard problem.

We wanted to be able to make it so we can use this device for our frequency response analysis. Yes, I don't know if you've seen that, but it means that we can do game face right? We do the same as the Bode 100. Basically right. Yep, with fully isolated chain but with fully isolated channels so you can go and and fully eyes to lay the signal generator.

So same thing. Yeah, that's right here. You do it in one place. you might as well work the other.

Yeah, so that means you can inject a signal into the error amplifier in your live power supply and measure live things. And the full frequency equals your 1500 volt. Exactly. I'll supply Wow That's right.

There is nothing on the market that does that except this. Quite right. Okay, so the we have these clock signals coming back so that the error between channels is about plus or minus 70 picoseconds. pretty small.

Hmm. I Wouldn't have thought that would be a necessary requirement on a four channel scope like this. But when you say frequency response analysis, well, you want very low phase zero? Yes, because if you want to measure things like inductance or capacitance, you need to have good, accurate phase. Yes! So the here are the eight there.

There are two lanes in each direction, so there are eight lanes coming in to the FPGA This is an area 5 FPGA so it has eight lanes coming into it. Crisi FPGAs It is a pricey Fpga variants. It's got a 960. There's a lot of walls.

That's right, it's A it's a very balls each other. Tell us about the choice of FPGA Did you need it for the you obviously didn't need it for the I/o count? I did Actually, because you might notice that this board plugs into this one and it's using in the equivalent of an Altaira daughterboard connector. Let's call it that. Oh yes.

okay, it's their standard connector. It's high bandwidth and we plan to offer more digitizer boards that you be able to plug in. and one of the digitizer boards we want to provide is a four channel one Giga sample per second board which is based on an on a A to Vicho. it means lots of parallel lines.

Lots of parallel lines, Exactly. No, we've got a lot of speed gates because it's upgradable in the field and we plan to add more value. Yes, it is. Yes.

Yep. So today and we have it supplying the application so you can just upgrade it as you go in the field for no extra cost. So then the other things down here. There's a power supply system, which there's lots of power supplies you need for an area V.

Sadly, yeah, you know, like 1 V 1, 1 B 1 V 1 V 2 1 V 8. Both of them not me. And this is our I Ord And we have an isolated signal generator Norther 65 mega Hertz And you can see here the two isolated transceivers that allow us to get signals across the gap. We're not actually using fiber for them.

These are pretty good. Okay, low currents it is. It's a 14 bit one and it samples out at a hundred and seventy mega samples, so it's a little bit slower and sampling out, but it still does a pretty good job it is. It's got a 4k arbitrary buffer so it's not huge, but it still can do quite a bit.

Is that done inside the FPGA? No. It's no. The problem we have is we it's isolated. No, we know we use an analog devices chip which has an arbitrary waveform generator built into it.

Got it? Yes, very handy and that gives us a frequency range of North Sixty five megahertz. Yeah with very low distortion. So the whole system is very low distortion. so we do better than 80 DB down which is not bad.

And then here is the USB 3 connection and we're using a nifty Di chip here for the for the for the USB 3 because it also supports USB 2. So you need USB 3 for the bandwidth. We we have a lot of data coming backwards and forwards. yep and we also do streaming.

So if you want to stream to disk you need you need high bandwidth. especially. You've got 4 channels at 14 but each. Does this have any local storage at all? Is there at 100 or is it like a hundred percent stream via USB No, the USB is far too slow for that, right? Okay, so I'm If you think about it, it's got 500 mega samples at 16 bits.

essentially. Okay, so that's like a Giga sample. There's four channels that's 4 times 16. Let's call that 2 bytes 4 times 2 bytes is 8 bytes times 500 mega samples.

It's four gigabytes per second. Oh, you cannot get 4 gigabytes per second across a USB link. but it's not being fired bytes and it has. It has 64 mega samples of storage.

All that we are and there are some on the other side as well. so we turn it over. we can see another two there. so that gives us 64 mega samples for every channel.

So this for megaforce channels of analog and one channel of digit 8 channels of digital. How does it? How long does it take to dump that full memory over to the PC if you use the entire acquisition memory it if you dumped a whole lot over USB 3 at USB 3, we're getting about 200 megabytes per second. Okay, so so but then we don't work that way. What we do is we just decimate our teenage samples at whatever Batwa rate you need to fit the Sebi anywhere we saw on the screen before we'll get in like a real-time update.

It's a long number of samples, Exactly right? Yes, But you can get the whole buffer into memory if you want. You just have to wait and and we stream at about 3 to 5 mega samples per second, 5 mega samples per second and then Dec can run for weeks if your hard drives big enough and we provide the tool so that you can zoom around and do a 50s and protocol decoding and metz and what's the stuff with the software? Can you? Does it support like big 4k screens and you can see that all the full 14 bit? Yeah, yeah, it's just not a problem. This is an EFP socket which stands for small form-factor socket and it's so that we can plug in an Ethernet module on there is like just an Ethernet because a lot of our users would like to have fiber. Oh, and instead of copper? Yes, and then if we you can see, these are the two sorts of modules.

So this is this is a wired one gigabit per second Ethernet module and this is a fiber One fiber is quite common you see and we would live to use up a lot more real estate to to put both in there. You obviously can't Even with one gigabit ethernet, you can't stream the fall. No, no. four gigabytes per second.

It's 32 gigabits per second. It's 32 gigabits per second. No, no even ting gigabits per second wouldn't do it for you who are your main customers for this sort of kid. Because it's not cheap, but you know it's $9,000 but it's the only thing on the market that does the job.

That's right. So our main customers to date are the people who want to who work. and Power Electronics and Power Electronics has becoming more more ubiquitous because it's used in electric cars. It's gonna say miserable manufacturers exactly.

So we we have. We have them at Bosch and and Siemens and and which we're trying to work with Delphi and Delco and the states who are also making you know engine controllers and things like that. So yeah, the people who use it are people are mostly into power electronics which is have the green tech movement which we see is that being a growing market? No you just we absolutely do. That's a general purpose scope.

it does all the same things. the frequency response analysis to streaming the mess, all that sort of stuff. but it's not as bandwidth, it's about $1,000 more. Yes, exactly the same software and you can link the two together if you really want to.

Some limitations but you know that's how it is. Yeah and this is been out for a while now but you still saw the hallways. Yeah yeah we this has been out for about a year and in the former turn but we are bringing out a new a Natale new board set which fixes app doesn't fix any serious deficiencies but make some things that we want to do bit a bit of any other design. Any other design gotchas on this that caused a real headache during development was the front end in is front end as a make front end is that easy peasy.

There's no, it's no it's that's that's probably why we're having a new board for the front end digitizer. One of the issues is that if you've got a 14 bit converter and your channel noises only 2 to LSPs So like if you're looking at 800 volts, you're getting 200 millivolts of channel noise. You really want to maintain that over the entire frequency range, don't you? You do? Okay now one of the big issues we have is probes. Probes themselves are not linear to that degree.

They might be good for 1% So one part in 100. We want one part in a thousand exactly. So that means if you're looking at a square wave, we want the top of it to be linear to within one. Pardon 8,000 Just Amalie.

Just just 200 millivolts actually what we're aiming for and 8/4 and 800 volts. sorry. And we found we could not get that with standard products. The solution is we're working with Pentek who make our products and we have modified the probes for them.

We have. We've changed their probe design. Well, they haven't done the design, we've done the tweaking and we're handing the design across to them to make. That's right, They hand it over their base design and you said well and you mucked around with it exactly.

And in the process of doing that, we also had to change up on front-end design. Our front end still works perfectly with the standard probe, but what we've done is we've matched the port the characteristic of the Pontic front end printing the probe with our front end so they beautifully match across the whole frequency bandwidth. And they give us one part in 8,000 linearity. What? So what did you have to tweak to do that and why don't they do it under standard? Well I Know why they don't do understand approach because they don't have to, they don't have to.

know. eight bit converters you know and you've probably got to knelt to bits of LSB noise as well. It's only one and 100 is all you need. And if I try and do these sorts of tests with my teeth scope that I can't even see it that you had to do.

or is that there was a secret secret? Square off? No, No. We ripped apart their their probes and and we discovered that we had to add in a couple of little extra capacitors and a little inductor to to match the poles and zeros that shouldn't have been there. And in our own front end, we found that when we switch range that we weren't having beautiful impedance, keeping the impedances exactly the same. So we have had an extra little switch that switches and a bit more impedance so that both channels exactly the same.

Got it Now Probe designs. Usually you can either get top probe compensated up here or compensated down there. Which one did you go with? This is compensated down here. it's isolated.

No, that's right. you can. You can. you have to undo that label on the label.

It's like this idea though. it's down there. No, it's got a little plastic huija me flickery so it's it's like that. except I've put down the label so that people don't turtle with it.

so you had to add the stuff to the probe inside the probe. No, it's actually doing the scene. This is where all the action happens. Yep, anything that PS there then contend with the impedance of the probe coax cable and it's much easier to do it.

and the scene. So that's where we've had to do it. That's right. And then our own changes are in the front end here.

Yep. Are they allowed to sell that probe to anyone else? Or is that your own custom? Well, they they could. but it's a little more expensive for them to make so and I doubt that they will because it won't match other people. Okay, this is a way being a probe.

Yeah, well. so a lot of the scopes are coming out as 10 or 12 bit now. So a lot of the high-end you know. Okay, Simon Texan right? Yeah, I know.

but they say they're 12 bit scopes. but if you go and look at the actual effective number of bits, it's still down at about 8. No, you know it's um, it's true there are 12 bits, but the noise are still pretty high. We are an order of magnitude bit of noise.

So yeah, that's life. So there any other really troublesome things that cost you hard HR in the development? Let me think their front in the front end. Yes, there have been lots of hassles in terms of components not working the way they should work. We started off life using well, we still are using some analog devices Op Amps and they speak for the Op-amp is that you can pare it down okay.

but when we powered it down and drag the input to minus five-fourths and we didn't expect that and neither did the designer who I spoke to an Ireland a woman who said that doesn't do that but then when she went and checked it the second problem that we had is is that it has a output that you are supposedly able to multiplex. So we have two ranges, the 800 millivolt, plus or minus 800 millivolt range and plus or minus eight ball range and we wanted the multiplex between the two ranges and we're going to use to our pumps and switch one off and switch to the other one on and I Talked to the designer and she said yes, this is going to work fine, you'll all be fine. We design it. Guess what? It wasn't all fine.

No there was leak through even when it was off so that there was a problem for us. So and then we've discovered the ADC that we use is actually inside it's two ADCs there 200 to 250 mega sampled a DC's and then to leave them they do and they have this engine. They called it an engine which which tries to get the gain and phase correct so that into the leaf beautifully. But it turns out that they did all their testing only in the RF domain on sinusoidal signals.

and when you put in a non sinusoidal Silas's signal non. Anyway, when you put in a single-ended signal that may be just the the bottom of an edge, the whole thing turns to custard. It took us a lot of effort to overcome that. We we've been working with Undersell to try and fix that.

They're not into Cilenti. No, they're now renesis. Yes, yes, they have. You're no longer.

But the guys inside are still then distill guys. Yeah, that's right. Yeah, so that was a little. Hesse Lee Yeah, How much total? How long did it take you all up to do this from go too long? Well, I reckon that's been about four years is just huge.

Yep, Yep. four years, three, two of them. Two of them would see that one of the guys is doing all the application software and the other one of the other guys is doing all the hardware and then I do so. FPGA and they see inside this thing here and I don't have full time available so no two and a half.

it's very nice to see you again.