Professional Scope Purchasing Criteria

[nctnico]

@TunerSandwich: IMHO you are going over the top. A good engineer knows what should be on the screen of the oscilloscope and for 99.9% of the measurements you don't need to get rid of noise and spikes.

I think you are under-estimating the gentlemans needs......and his very specific applications.  I would re-read his specific requirements.  Also...how can you go "over the top" in scientific measurement? 

If you connect a standard probe to a board with a hefty DC-DC convert to check an SPI interfave you'll likely pickup some noise from the DC-DC converter. As long as that noise isn't in the way of getting the information you want there is no need to spend a lot of time or money on trying to get the signal to look nicer on your oscilloscope. At some point you simply have enough information.

[rx8pilot]

At some point you simply have enough information.

Agreed, the question is how much information is enough to make decisions and sleep well at night - at least for me. In other areas of my professional career, I am an 'expert' and able to infer certain results based on experience to determine whether all is good or not while only having partial information. My experience in designing electronics is only a few years deep so it is harder to make a call about whether noise or spikes are ok or not. So far, my more common experience is seeing noise or spikes on the scope and mistakenly chasing down a circuit problem only to find later that the test was creating the problem. I have dedicated some time to understanding the combination of technique and tools needed to really optimize the designs as opposed to 'just getting it to work'. In the end the real world will always prevail - I can't afford a 1Ghz scope with $10k worth of probes added in and isolate my lab in a Faraday cage. I also don't have months to measure and optimize a simple-ish circuit. I need to pick some reasonable starting point that will offer the least amount of limitations for a reasonable time. Obviously, a lot of variables presented there - so I have to take a leap of faith to some extent.

A "good engineer" is one who takes into account ALL of the available data, and uses the tools at his/her disposal to understand where things can be improved.

Nice.
To add to that.... I take all of the the RELEVANT data and divide by the time allowed, sort problems by order of value to the project and start solving. I have worked with and paid engineers that take all the data available and it simply baffles them to the point of being frozen. Knowing what needs to be solved and in what order is a critical engineering skill that does not come with all engineers. For my business to work, I have to keep a high-altitude view of the project and then swoop into the details to do the actual engineering. Spend too much time in the details and you lose track of what you are trying to accomplish.

I am planning to get the Agilent 34461A 6.5 digit DMM and a multi channel electronic load system [probably BK since the prices and performance are reasonable for multi-channel]. This will allow some automation in the testing which will same some time.

Why would I need a voltage reference? Do you see that as critical?

Also, I am putting the lab in separate building from the CNC shop. Not sure what kind of mayhem they spew into the electrical - never measured or even thought of it. With 30Hp spindles starting and stopping along with huge servos that move an 800 pound table at 1,400 in/minute stop-start, it could be nuts.

Looking forward to seeing the screen grabs. Also thinking more and more about renting a 200Mhz and maybe a 500Mhz or 1Ghz together for a month to understand how they will impact the daily challenges. I am lucky enough that a rental company is about 3 miles from my house. Being able to compare with my actual circuits on my bench could reveal a LOT about what is best. It should also teach me a LOT about technique using high-end instruments.

[LabSpokane]

Maybe you should spend less on a scope now and invest some into a separate step down transformer off your mains. Trying to test while fighting crappy power is un-fun.  It's a great opportunity to have a few new ground rods pounded in anyway. 

[rx8pilot]

No problem - separated from CNC by 12 miles. Should be reasonable HF fall-off at that distance 

[TunerSandwich]

@TunerSandwich: IMHO you are going over the top. A good engineer knows what should be on the screen of the oscilloscope and for 99.9% of the measurements you don't need to get rid of noise and spikes.

I think you are under-estimating the gentlemans needs......and his very specific applications.  I would re-read his specific requirements.  Also...how can you go "over the top" in scientific measurement? 

If you connect a standard probe to a board with a hefty DC-DC convert to check an SPI interfave you'll likely pickup some noise from the DC-DC converter. As long as that noise isn't in the way of getting the information you want there is no need to spend a lot of time or money on trying to get the signal to look nicer on your oscilloscope. At some point you simply have enough information.

That is making a HUGE assumption, that the DC POL is in fact stable and doing what it's supposed to....in relation to the dynamic load being presented.  You are missing a massive chunk of the OP's equation.  You can't assume that the stability of a digitally controlled POL is going to be acceptable, for driving an FPGA/ASIC or similar.  The OP's use case scenario is far more complex than you are allowing for.  He didn't say he was whacking a beefy DC-DC reg on the front of something.  He specifically said he was dealing with a multi-phase architecture. 

ALSO, by your own admission, you say that the use case scenario I presented isn't relevant 99.9% of the time.  Well Murphy will make sure that the 0.1% of the time happens 100% of the time 

Again, i am not sure the nature of multi-phase DC,POL solutions is being understood here.  The step response of those devices is absolutely critical in gaining stability....especially for complex multi-phase IBV systems....when the load is dynamic and one would be concerned over dead times and V-sense. 

We aren't talking about a 5v linear DC regulator here.  We are talking about control over a multi=phase architecture, that employs POL changeovers, being controlled by sensor feedback into a MCU or similar.  These are the devices I deal with on a daily basis, and I can assure you that these specific topologies are nowhere near as simple as you are making them out to be. 

I can also assure, that my experience in this VERY specific field, has taught me that 100MHz front ends, with low sample rates is going to turn over and bite you in the ass, when you least expect it.  Sure everything will be just fine 90% of the time, and meet your engineering predictions.....but that's not engineering.  Any idiot can follow a build example on a datasheet.....when you really earn your pay is when things go the opposite of "plans".  If you have to hire out, because of the lack of insight, in your own T&M setup....then the extra $10k you were questioning spending....becomes a VERY VERY valuable investment.  Been there many times. 

One guarantee we all have, is that nothing will ever go 100% according to plans.....you can bank on that.  Hence the reason for test and measurement gear, to begin with.  If everything stayed perfect 100% of the time, we wouldn't need to check.  It's also complacent to think that the laws of physics are fully understood.....or that some joker at the supplier didn't tag a 22uF cap as a 47uF cap. 

Shit happens 

[tautech]

@TunerSandwich
As you seem to have a pretty good handle on the OP's requirements, and the experience to boot, what would be the minimum specs you would see as non-negotiable for his needs ?
BW, Sampling rate, Memory, I/O's, Decoding...etc.

[TunerSandwich]

At some point you simply have enough information.

Agreed, the question is how much information is enough to make decisions and sleep well at night - at least for me. In other areas of my professional career, I am an 'expert' and able to infer certain results based on experience to determine whether all is good or not while only having partial information. My experience in designing electronics is only a few years deep so it is harder to make a call about whether noise or spikes are ok or not. So far, my more common experience is seeing noise or spikes on the scope and mistakenly chasing down a circuit problem only to find later that the test was creating the problem. I have dedicated some time to understanding the combination of technique and tools needed to really optimize the designs as opposed to 'just getting it to work'. In the end the real world will always prevail - I can't afford a 1Ghz scope with $10k worth of probes added in and isolate my lab in a Faraday cage. I also don't have months to measure and optimize a simple-ish circuit. I need to pick some reasonable starting point that will offer the least amount of limitations for a reasonable time. Obviously, a lot of variables presented there - so I have to take a leap of faith to some extent.

A "good engineer" is one who takes into account ALL of the available data, and uses the tools at his/her disposal to understand where things can be improved.

Nice.
To add to that.... I take all of the the RELEVANT data and divide by the time allowed, sort problems by order of value to the project and start solving. I have worked with and paid engineers that take all the data available and it simply baffles them to the point of being frozen. Knowing what needs to be solved and in what order is a critical engineering skill that does not come with all engineers. For my business to work, I have to keep a high-altitude view of the project and then swoop into the details to do the actual engineering. Spend too much time in the details and you lose track of what you are trying to accomplish.

I am planning to get the Agilent 34461A 6.5 digit DMM and a multi channel electronic load system [probably BK since the prices and performance are reasonable for multi-channel]. This will allow some automation in the testing which will same some time.

Why would I need a voltage reference? Do you see that as critical?

Also, I am putting the lab in separate building from the CNC shop. Not sure what kind of mayhem they spew into the electrical - never measured or even thought of it. With 30Hp spindles starting and stopping along with huge servos that move an 800 pound table at 1,400 in/minute stop-start, it could be nuts.

Looking forward to seeing the screen grabs. Also thinking more and more about renting a 200Mhz and maybe a 500Mhz or 1Ghz together for a month to understand how they will impact the daily challenges. I am lucky enough that a rental company is about 3 miles from my house. Being able to compare with my actual circuits on my bench could reveal a LOT about what is best. It should also teach me a LOT about technique using high-end instruments.

Your experience and path, very much mirrors mine.  I am not only the lead developer/engineer at my company, but i also own the company....and i have to put on different hats (as you do) when it's necessary.

Obviously we NEVER want to get caught up in endless scope crawl.  Also, OF COURSE, we only want ot look at relevant data.  However, what is relevant isn't always obvious.  It's always the most seemingly minor or irrelevant thing that comes back to bite you in the ass.  Without fail.  It's much better to have over=provisioned and "over engineered" something....than to have to deal with the bad will of a recall.  Or discover a week before a product ships that you have some horrific instability or safety issue. 

Hell even the best engineering firms have bodges on their boards.  So we can guarantee that, at some point someone in the production and dev chain will be complacent....or simply make a bad call. 

There is a very fine line between having all the facts, picking the most efficient and cost effective route and saving a buck.  It's not easy.  I am sure you know that intimately well.

At the same time getting caught up in endless datasheet and equipment marketing is also no solution.  As with all things it's specif and has to be balanced (pun on differential pairs there?).

However, banking on the experience of others, and using your own educated judgement is a recipe for success.  If you can't trust the opinions or work of the people you are hiring, as engineers, and feel you have to play stunt double for them....well then you really have to question why you are keeping them around....?  Trust me, i know how difficult it is to walk this line, but at some point you have to be realistic about reality.

The worst thing you could wind up doing id buying into a system that doesn't do what you need it to.  It would have been far better to lease, or hire out a lab with the correct tools.  I get that budgets dictate reality, but so do results.  You can't pick one over the other right?  If you spend half of what you should have on T&M, AND the product flops because of design problems....well it would have been better to not spend anything at all.

I am not here trying to say you have to spend half a million dollars on absolute bleeding edge systems....at the same time, the fact that you are questioning what you really need....kind of says that it's impossible to grasp the reality of a budget.  I mean you don't define budgets that way.  You don't say I have $xxx to spend, so let me figure out what I need.  Instead you say, let me figure out what i really need, and then how to pay for it.  Sometimes allowing a pre-conception to dictate a budget is a foundation for disaster (been there many many many many many times). 

I think it's best to buy this equipment, when you have enough insight to make confident decisions based on your own knowledge.  Obviously no-one here can tell you "without doubt this is what you need".  I am personally just trying to look at your very specific use-case scenario and use relevant experience to give insight on what has worked for me.  Does that mean you MUST have xyz or ABC?  Of course not.....and hey you might get really lucky and everything goes off w/o a hitch.....but you must provision for the opposite, so that it doesn't take the entire ship to the depths of the sea.....

Glad to hear the lab will be offsite.....the garbage that big boy machinery spews into the building electrical system, and radiates through the "ether" is a hell I wish upon no-one.  I have spent more money than I care to mention, trying to band-aid those kind of problems, and it's always a game of compromise.  In the end you wind up spending more on mu-metal, isolation transformers and chemical grounds than you would have just building a new structure, from scratch, off-site.  YIKES!!!

The voltage reference and calibration refs are critical, because you mentioned using a 16 bit external A/D.  You will have to lock that to an external V-ref.  For example a 12 bit A/D would need a 4.096V A-ref to equal a logical float, in programming.  I could get into this more if you don't already know on the traps that crop up there. 

Now comes the fun part.  Lets say you do need to employ this precision V-ref.  The V-ref will need to be calibrated, per device.  To do that you first have to trust that your calibration meterS (notice the emphasis on plurality) and in fact trustworthy.  Also that level of traceability needs to be ongoing.  So ONE quality, calibrated 6 1/2 digit MM is simply not enough to bank on.  You need a minimum of two, on the same test point....both with identical (as close as possible) test leads etc....BOTH of those units need to be warmed up and verified against a known, traceable reference....I would strongly advise budgeting for that type of situation.

The architecture you explained earlier in the thread, is NOT some basic/simple topology.  Multi-phase power solutions, driving highly sensitive processing blocks are MUCH more complex and the precision required to keep them working and warrantied is far from "well that's good enough".....

I will set-up a test tonight or tomorrow, and show you some differences between a variety of solutions...and will purposely inject a fault into the topology....so you can see the 100MHz DS (Rigol) totally miss it.....hopefully a few simple screenshots will illustrate my points here, but if not then I could possibly make a video.....

I have been at this specific engineering discipline for close to 20 years now, and believe me i have made ALL the mistakes....and sometimes they have cost me dearly....I am not trying to present doom and gloom, and say you will fail if blah blah blah....obviously everything is specific and blanket statements are utterly worthless.....but that is the point here....we are getting very specific....

[rx8pilot]

There is a very fine line between having all the facts, picking the most efficient and cost effective route and saving a buck.  It's not easy.  I am sure you know that intimately well.

Saving a buck is the most expensive thing I have ever endeavored to do.

I am not here trying to say you have to spend half a million dollars on absolute bleeding edge systems....at the same time, the fact that you are questioning what you really need....kind of says that it's impossible to grasp the reality of a budget.  I mean you don't define budgets that way.  You don't say I have $xxx to spend, so let me figure out what I need.  Instead you say, let me figure out what i really need, and then how to pay for it.  Sometimes allowing a pre-conception to dictate a budget is a foundation for disaster (been there many many many many many times).

Past experience has taught me the very hard and expensive lesson of buying more than you need is cheaper than buying less and failing because of it. It certainly helps that Agilent has 500Mhz MSOX3054A for $6k! Even the options are discounted. That is only $1k above my original thought of spending about $3-5k and I would not likely outgrow that. The Tek MDO3k looks great, but I have not seen one at nearly 50% off either.

The architecture you explained earlier in the thread, is NOT some basic/simple topology.  Multi-phase power solutions, driving highly sensitive processing blocks are MUCH more complex and the precision required to keep them working and warrantied is far from "well that's good enough".....

I wish I had this conversation a year ago when I started going down this path. The learning curve has been outrageous and I am just getting up to the starting blocks. It seems like balancing a bowling ball on top of a nail sometimes. Converters aside....the challenge of switching sources is not trivial . This device has three inputs that are prioritized, when one droops or fails it switches to the next one. Challenge #1 is that I do not control the supplied power. It could be a battery of any chemistry, a switching supply of any quality AND it can be supplied from an unknown quality of cable of varying lengths. That all by itself is difficult. It must react VERY fast but cannot tolerate false sensing during a transient load. The loads are also out of my control so I have be prepared for all sorts of scenarios. Some have tons of isolation that makes life easy and others will fall to pieces with the slightest glitch. Also, when one output is shorted or starting a big capacitive load, the rest of the outputs have to remain stable.  On top of all that, all inputs have to deal with revers polarity in a way that it will not power a load, but power the logic so the display can inform the user of the condition. I really had no idea what I was getting myself into, but now that I am here, it's great to have the understanding of how much I don't know.

The biggest driver of all the upgrade is to improve the performance and reliability of this system. I really appreciate the time you have spent participating in this discussion. The insight from someone doing the same thing, but with far more experience, is extremely valuable. Thank you.

[TunerSandwich]

@TunerSandwich
As you seem to have a pretty good handle on the OP's requirements, and the experience to boot, what would be the minimum specs you would see as non-negotiable for his needs ?
BW, Sampling rate, Memory, I/O's, Decoding...etc.

That really is the million dollar question isn't it?

It's a very difficult one to answer, without throwing out blanket statements and generalizations.  Last time I checked critical systems didn't work on assumptions and guesses. 

I think it's near impossible to answer that truthfully, without knowing more specific details, and possibly seeing a schematic. 

The only honest answer I can come up with is "as much as he can afford"....

I know for the systems I am working with (very similar), 300-400 MHz would be a minimum.....a sample rate fast enough to give me FFT results out to what my physical capacitance constraints dictate....a probing solution that doesn't inject error into my measurement, with enough bandwidth to really see resonance faults (especially in the feedback loop of the POL v-sense)...memory depth should be enough so that my scope doesn't go into roll mode at the longest time division I need to measure (that is a bit of a misleading answer) and allows me to zoom in and resolve runt/overshoot errors...I/o's should be a minimum of 4, and the more the better.....logic probes/analyzers are less useful than full BW channels....unless we are talking about massive parallel buss (and we aren't)...

so what is MY ideal scope?  hell that's a good one....let's dream a bit here....

1. 3GHz 8 channel front end (minimum 12 bit vertical resolution)
2. Minimum 2-5 GS per channel....more is nice =)
3. fast Agilent style waveform update rate...
4. better than 14 Mpts (hopefully a lot better)
5. runs windows OS, for iScsi and development ability
6. has a plethora of math and decoding options
7. support for 10 years, and free access to firmware and tech documentation
8.  the ability to buy service parts direct, for a reasonable cost....
9.  represents a phenomenal value that amortizes across the entire life of the hardware...
10. LOCKS TO EXTERNAL SAMPLE CLOCK!!!  and distributes unaltered clock to other devices....

yup I know, it doesn't exist =)

[TunerSandwich]

I wish I had this conversation a year ago when I started going down this path. The learning curve has been outrageous and I am just getting up to the starting blocks. It seems like balancing a bowling ball on top of a nail sometimes. Converters aside....the challenge of switching sources is not trivial . This device has three inputs that are prioritized, when one droops or fails it switches to the next one. Challenge #1 is that I do not control the supplied power. It could be a battery of any chemistry, a switching supply of any quality AND it can be supplied from an unknown quality of cable of varying lengths. That all by itself is difficult. It must react VERY fast but cannot tolerate false sensing during a transient load. The loads are also out of my control so I have be prepared for all sorts of scenarios. Some have tons of isolation that makes life easy and others will fall to pieces with the slightest glitch. Also, when one output is shorted or starting a big capacitive load, the rest of the outputs have to remain stable.  On top of all that, all inputs have to deal with revers polarity in a way that it will not power a load, but power the logic so the display can inform the user of the condition. I really had no idea what I was getting myself into, but now that I am here, it's great to have the understanding of how much I don't know.

The biggest driver of all the upgrade is to improve the performance and reliability of this system. I really appreciate the time you have spent participating in this discussion. The insight from someone doing the same thing, but with far more experience, is extremely valuable. Thank you.

I work on those EXACT types of systems....but my dynamic loads are resistance wire....with wildly varying tempco....and then the secondary loads are processing systems, that need to process sensor feedback on the primary load...EXTREMELY fast.  We need power accuracies in our systems, that really even skirt the usefulness of 6 1/2 dig DMM's.  In some cases extremely tight clocks and "over the top" VR is required. 

The other big problems in these switchers is ringing in the inductor (physical).  Especially with a low profile SMT solution.  If the thermal constraints are bonded to the PCB...then this ringing can cause havoc to multilayer ceramic caps.  As you well know we NEED ceramics for their low ESR (on sense loops especially) and big electrolytic is a no no....for both thermal AND freq vs impedance issues.  Also tantalum  is a no go in fire rated systems.  No way am I sticking a 22uF tantalum on a system with a dynamic or shorting load.....so we are slaves to the ceramics....and the number one cause of instability in DC pol regulation is ringing in the caps/feedback loop...ESPECIALLY if you reference v-sense to the control loop....

I can tell you FOR SURE, you are going to encounter this set of problems....especially when your PCB density goes up...and that happens AFTER prototyping.  Then shit gets crazy (crazy fun with the right tools) 

You DO need a device that can resolve these problems.  Can you solve them with pencil and paper....YOU BET....and by the time you use that method + trial and error, the market will have abandoned that technology and you will be an old man 

Modern scopes are here to HELP us get things to market.  That is the trend is ALL development tools these days.  Time to market is everything today.  That isn't going to change.  This isn't 1950, where a TV repair man comes and plays with some tubes and guesses around about some unknown physics.  We understand more about our world today, than at any other point in history.  The time to guess is well over. 

That is what these modern tools are about.  It's not enough to see a sine wave at 20 MHz anymore.  That is the realm of hobby/maker discussion.  If you want to be serious and competitive and move with the times then you need to employ tools that HELP you do that.  Looking at some monochrome screen and saying "yeah it's noise, but that must just be the probe" is not going to go very far these days.

Even in government contracting it's getting serious (is that even possible?).  The public sector is looking to the private sector....for BIG things....hell even space travel is becoming "commoditzed".  Gone are the days of bloated 20 year projects.  The resources just aren't there anymore.  Financial, physical and intellectual. 

I never understood why so many engineering students get stuck in "teaching"....not until very recently.  Traditional tools and programs just don't prepare people for the reality of the industrial machine.  About all a traditional engineering degree and toolset is good for these days, is pulling your dick under the podium of some theoretical physics lecture....in the meantime people are actually making things that dramatically CHANGE humanity and our needs/wants. 

I know that's a bit far off oscilloscope discussion, but let me tie it together. 

With these modern ideas, systems and economical approaches...it's no longer good enough to get by with "just good enough".  Before I am dead I will have 9 billion minds to compete with...for my share of the remaining resources.  I sure as hell want every edge possible to get there.  If that means i have to look at things a new way....and drop my per-conceptions about "only ever needing xyz bandwidth"....or "eh don't worry about that glitch...it still works"....SO BE IT!!!

If you would have asked me what tools I needed 5 years ago, to "make something happen", I surer than shit wouldn't have said what i am saying today.  Such is the pace of "progress".

If you ask me what I need today, to look out to TOMORROW, well then I have learned some very valuable lessons.....and my answers would be a lot more open minded and challenging than "I need this"....

The real answer to your criteria, is get what you need to realize ALL of the potential problems...and potential diversification benefits of doing something "new".  If your technology is refined enough and versatile enough you can re-purpose it through endless re-iterations (more like repackagings) and absorb the investment in those "fancier" tools across a more realistic product cycle life.  Hell it has worked brilliantly for apple and their little portable phone thingy.  Same damn thing, in 6 plus re-packagings...and idiot masses still MUST HAVE IT!!!  They invested MORE up front...to do something really cutting edge, so that they could reap the ever increasing margins on re-selling the same thing 10 years later. 

Sounds like a winning plan to me....seems to have worked for them.  That philosophy, coupled with good marketing...has allowed an electronics company (don't forget) to take over the mobile communications AND music industries....

If your product is successful....you won't be worried about an extra $10k you spent at the beginning....but if it fails, you will always wonder if/why you didn't buy the right tools to lead to success.

Damn that is one annoyingly long winded response....sorry about that

[tautech]

Damn that is one annoyingly long winded response....sorry about that

Minister TunerSandwich stands at the pulpit and delivers his sermon.   

Wonderful to see such passion in this profession. It is a rare thing.

Meanwhile tautech goes looking for the previously spec'ed scope.......................

[nctnico]

At some point you simply have enough information.

Agreed, the question is how much information is enough to make decisions and sleep well at night - at least for me. In other areas of my professional career, I am an 'expert' and able to infer certain results based on experience to determine whether all is good or not while only having partial information. My experience in designing electronics is only a few years deep so it is harder to make a call about whether noise or spikes are ok or not. So far, my more common experience is seeing noise or spikes on the scope and mistakenly chasing down a circuit problem only to find later that the test was creating the problem.

The statement above makes me wonder: do you simulate your circuits? Many of my more complex circuits start as a simulation to get a feel on how a circuit will behave, what waveforms to expect and which parameters have the biggest influence (power supply, temperature, etc). Sometimes I try various solutions in a simulation before building a prototype. In a simulated circuit it is also much easier to look at the current through several components. The downside is that simulating a circuit and getting meaningfull information is a bit of an art in itself.

[rx8pilot]

I have not simulated any of the circuits. Its hard to tell whether its a worthwhile skill/exercise compared to "buildit and study". My ME experiences with simulation did not serve me well. I believe that is mainly because I could physically build a machine and go through a myriad of tests to destruction to build design performance data. the time it would take to build a good simulation and study it was almost always longer and the resulting data (even if perfect) was limited. Building a prototype gave us information that a simulation would never deliver. Now, keep in mind, I do not design or build systems for flight, medical, etc that require far more testing and documentation. Also, just one parameter off and your data looks great, but is trash and you don't find out until you build it for real anyway.

My electronics adventures seem to be similar, except that I don't know anything about simulation or the data I could get from it. On paper, it seems fantastic and almost mandatory. in reality, I fear that I would spend a ton of time and get only a handful of useful data. I really don't know - just a guess. It seems that what I struggle with in general is very subtle nuances that a simulation would not help unless I spent a huge amount of time putting it together. I can layout a PCB and have prototype very quickly and I have the equipment to build with even the finest components. My thinking is that if I go through that, I am not only testing the schematic, but also the PCB layout. The layout, in my case, can easily be the difference between a perfect result and a mysterious unstable mess [probably true for a lot of designs]. Do you think this could be a logical fallacy? I am projecting bad experiences from an unrelated discipline here, so I don't really know.

Like I said, it seems great to know the 'ideal' result to be looking for. From one end to the other, my current project has a little over 500 components and 180 BOM line items. I would be totally intimidated with creating a simulation of that for sure. I could have simulated smaller sections, but overall it would take the same amount of time I guess. I will read up on the subject to get an overview of what it involves, software available, required skills, etc...

[nctnico]

I think simulation is definitely something which could aid you to help understand what is on the screen of your oscilloscope.

During my EE study there was a lot of focus on simulating circuits. In one lab project we had to calculate a circuit 'by hand', simulate the circuit and built it. This was an interesting exercise but I didn't feel I had learned enough about simulating a circuit so I took on a much larger project: designing, simulating and building a pre-regulated 600W linear lab power supply with MOSFETs. The simulation quickly showed some things I overlooked which I fixed in the simulation. When I build the power supply I also discovered that some parts didn't behave as expected. It turned out the simulated circuit behaved the same but that I just didn't test those conditions. Nowadays I often go back and forth between a real circuit and the simulation results to check whether what I'm seeing on my oscilloscope is right or not.

Simulating an entire circuit is a good exercise to learn what is useful to simulate and what does not need to be simulated. If you get more grip on simulating a circuit you will be able to create simplified models of the important parts of your circuits. Simulation can also help to find an elusive problem in a circuit: A couple of years ago I had to get a device through an EMC test and I only had 2 days to solve several issues. One problem area was an isolated DC-DC converter. The HF interference caused be the MOSFET was easy to solve by adding some gate resistance and getting the snubber right. But it turned out there was another source of HF ringing as well at the secondary side. The board was quite crowded and I didn't had enough time and boards to try several different methods in a real circuit. I also needed a deeper insight on what was going on in the circuit so I created a simplified model of the DC-DC converter which showed the same ringing as the real circuit. By looking at the current through the components I found out the reverse recovery of the rectifier diode was causing a nasty spike. I simulated several possible solutions and ended up with putting a ferrite bead in series with the diode. Needless to say the board passed the EMC test.

[coppice]

Many of my more complex circuits start as a simulation to get a feel on how a circuit will behave, what waveforms to expect and which parameters have the biggest influence (power supply, temperature, etc). Sometimes I try various solutions in a simulation before building a prototype. In a simulated circuit it is also much easier to look at the current through several components. The downside is that simulating a circuit and getting meaningful information is a bit of an art in itself.

The beauty of getting information out of a simulation is your probes do not influence the circuit at all.  Anything you see on a scope must be treated with suspicion.

[rx8pilot]

Anything you see on a scope must be treated with suspicion.

The truth.

I think simulation is definitely something which could aid you to help understand what is on the screen of your oscilloscope.

Any recommendations on where to start? Software? Educational resources? I can see the wisdom and will dedicate some time to the subject. Do you find it a challenge to build the models of specialty silicon or is it SOP for the manufactures to provide simulation models?

[TunerSandwich]

Simulations are only useful for very basic predictions.  They mostly represent static ideals.  Also the simulations have to be properly characterized for the specific components you are using.  It's virtually impossible for a simulator to predict something like a resonant interaction between a PCB and a component, or a thermal condition which isn't yet characterized by the simulator. 

They are certainly helpful tools, but are no substitute for probing a real production PCB.  In some ways they can lend false security.  Even the best simulators (dassault) fall short of 100% prediction. 

Let's look at some simulations vs real world response of a dPOL signal path.  This is a simulation of AC response.  Freq vs energy/phase.....I can do step response later this week (when I have some time to set-up the experiment).

[TunerSandwich]

In the simulation you can clearly see a predicted response curve.....which is partially represented in the real wold....BUT have a look at the HF energy which the simulation missed.

This problem is compounded with a dynamic load (I will simulate that on the next experiment).

I purposely use crap power supplies for these tests....to see how input phase/ripple changes the dynamic responses of the POL under test.  It's great to have a collection of crap SMPS and linear PSU's.  Feeding precise, filtered signal into the POL front end is a terrible way to test these devices.  In the real world we have to assume that our hardware is being connected to crap wall-warts or similar.  I can also run this experiment using a battery, with both high and low margining.  You will see just how far off the predictions are from the real world response

[rx8pilot]

This is interesting, although I don't feel I am able to understand exactly what is going on. I will look it over one more time and probably have some questions...

Thank you for the setup, hoping to be able to interpret what is going on.

[nctnico]

In the simulation you can clearly see a predicted response curve.....which is partially represented in the real wold....BUT have a look at the HF energy which the simulation missed.

This problem is compounded with a dynamic load (I will simulate that on the next experiment).

I purposely use crap power supplies for these tests....to see how input phase/ripple changes the dynamic responses of the POL under test.  It's great to have a collection of crap SMPS and linear PSU's.  Feeding precise, filtered signal into the POL front end is a terrible way to test these devices.  In the real world we have to assume that our hardware is being connected to crap wall-warts or similar.  I can also run this experiment using a battery, with both high and low margining.  You will see just how far off the predictions are from the real world response

IMHO that is due to the limited model you are using for the simulation. One of the things I like to simulate are worst case conditions which often reveal hidden problems. A 'crappy SMPS' is not very scientific 

[rx8pilot]

While not true in all cases, my experience in ME was that by the time I created a suitable model for a simulation that was to be believable - I could have built and destructive tested it a few times already which gives me the ultimate answer. Even after a good simulation has been built and analyzed, you still have to built and delicately test the circuit to confirm and look for anything unexpected. If there is any truth in that - why/when bother with simulation?

I too test with "crappy SMPS" sources to learn how my circuit deals with a variety of non-ideal situations. It seems that simulating and scientific analysis [guessing] would not be nearly as good as throwing the real world at it, right? I have not spent much time understanding my own definition of 'crappy' simply because my customers wont or cannot adhere to that - if my device says '12v IN' and their random SMPS says '12V OUT' - they will simply plug it in and expect it to work. I can setup a test with a variety of DC sources and examine the resulting performance impact on my circuit.

Is this a logical fallacy, or a reasonable way to go about it? Maybe use a simulation to set a baseline to measure from?

[nctnico]

While not true in all cases, my experience in ME was that by the time I created a suitable model for a simulation that was to be believable - I could have built and destructive tested it a few times already which gives me the ultimate answer. Even after a good simulation has been built and analyzed, you still have to built and delicately test the circuit to confirm and look for anything unexpected. If there is any truth in that - why/when bother with simulation?

A simulation can show whether a circuit idea can work. Simulating for worst case conditions can show whether a circuit works in real life. In a simulation you can check the effect of a crappy power supply on the control loop. You can test several control loop methodologies to see which one is best versus the number of components required. Again, it took me quite an investment in time to learn how to simulate circuits and interpret the results but I do feel it was worth it.

[rx8pilot]

Again, it took me quite an investment in time to learn how to simulate circuits and interpret the results but I do feel it was worth it.

What? You need skills for this stuff? 

Seriously....in ballpark terms, can you say what it took financially and temporally to get a workable simulation skill set. I know there are a million variables in that question, so maybe just a note on your own experience. I am over committed at the moment on education so I have to choose my battles wisely. The sequence in which I learn is nearly as important as what I learn.

[nctnico]

Though question. I'd start by working through a small (online?) course on how to use Pspice. Perhaps spend two hours per week. That should give you a good grip on 'the tool'. The next step would be to take on a project to see where simulation diverts reality and why. There are also mathematical pitfalls causing the simulator to quit. Sometimes it takes placing some extra resistors to get a circuit to simulate. A bridge rectifier is such an example.

[TunerSandwich]

This is interesting, although I don't feel I am able to understand exactly what is going on. I will look it over one more time and probably have some questions...

Thank you for the setup, hoping to be able to interpret what is going on.

Let's break it down. 

First let's understand the signal path.  Leading up to the probe test points. 

The chain starts with with a Korad KA3005D.  I chose this supply, because it has two criteria that are fundamentally misleading.  Number one it consistently has an overshoot of about 100mA beyond it's current limit set-point.  Number two, despite being a "linear" supply it injects about 50mV of "ripple/noise" into the IBV of the circuit under test.  The next point is it's less than ideal regulation.  It seems to not be able to handle transient events, with much elegance.  We could do a separate review of this specific unit, but I believe others have done so already. 

The PSU is routed to the next item through a "purposely dodgy" set of test leads.  They are stranded wire, of unknown origin, and are terminated into the PSU with alligator clips.  Certainly not an ideal connection. 

The other side of the leads are terminated with screw down, banana style terminations.  Those are coupled into female banana receptacles of "high quality" (Pomona electric). 

The female banana couples to a deans connector, using solid core OFC wire (very high quality test lead/adapter, I built in house). 

The next piece of hardware in the chain is a "hobby grade" power measurement device, from "astro flight" (not sure what pcb is being repackaged inside).  That piece of hardware shows me the drop in the leads and shows watt hours used.  It draws around 50mA of parasitic current.  Again I not trying to build the "best possible" front end power, leading to IBV.

From the power meter we again have well made deans to banana pigtail (overkill quality, built in house).

That is where the external power ends and merges into the IBV section of the circuit in test. 

The circuit in test is quite complex, and I won't go into lengthy detail, but it's basically a single phase dPOL and controller (based on an atmel MCU, but with high speed acquisition and switching control topology).

The simulation is the programming interface for the dPOL and controller.  There is no better simulation available, as it is built specifically for this circuit topology. 

In the programming interface I have control over an internal feedback loop/filter on the POL itself, as well as the sequencing signal sent (via 56 bit com signal) directly to the POL.

The output has been optimized for the "best case scenario".  Margining is set up, so is tracking and so is ultra tight VR.  Our output voltage on the pol is set to 5.5 volts.

The output V sense is coupled back to the digital filter section/feedback loop.  So the meters don't load the front end of the pol AT ALL.

We can clearly see that two calibrated and verified meters are reading very close to 5.5 V.  One is set with fast integration time, the other is not.  That is a point that should not be overlooked. 

The test leads are of high quality and free of faults (the main DMM test lead is shielded and properly terminated into 10Gohm). 

The scope probes are active FET units and have been fully deskewed, compensated and temperature calibrated.  They have also been nulled against a direct ground, to verify that no noise is being injected in the HF ranges.  They are 1Mohm 1.8pF with a ratio of 10:1 and a dynamic range of 5v....the dynamic range is matched to scope input rise time quite well. 

So that is a pretty ideal test right?  There is a trap though....I was hoping someone would spot it, but maybe it's a bit too complex to be obvious.

Pay careful attention to the voltage min max during the switching pulse.  I am triggering off of that event on the scope.  Have a look at the readings.  We are def overshooting 5.5 V, on both the positive and negative edges of the full switching phase. 

There is NO bandwidth limiting on the scope front end.  If I was to do some sin x/x interpolation, some averaging and employ a 20MHz input filter (AC coupled) we would completely miss this HF overshoot.

So let me explain the trap now.  The POL output has 2 4.7uF caps to help it out.  Obviously those caps are ballasted and properly used.  They are ultra low ESR multilayer ceramic. 

The sense loop has been calibrated to respond out to around 10KHz.  OOPS!!!!  There should be another set of parallel filters on that control loop!!!  I posted the response graph earlier in this thread.  So essentially there is a physical ringing external to the POL, but coupled into the control loop.  That is why we are overshooting our 5.5 v target, during the POL switching step response

I can go into more detail here, as this is obviously a dramatically over-simplified explanation.....but suffice it to say that we wouldn't have caught these problems on the REAL output of the device, if we hadn't employed this "overkill" methodology of measurement. 

Is that peak voltage enough to cause an issue in the turn on sequence of a processor?  Dunno....that depends on the processor, application and target use.  I know that it's an UN-acceptable scenario in my book.  We are overshooting 0.05044 V above our target. 

My point here isn't to explain WHY this is happening or how to solve it....the point is simply to illustrate that the simulation fell short of predicting the real world response and that real world response would have been missed with tools of lower bandwidth