Widebands That You Would or Wouldn't Buy

[toalan]

Unsure if I already posted this, but I (with help/others did this) correctly wired up a Spartan on the WikiSpeed car and got a nice noise-free signal from it. Obviously I can't vouch for the accuracy, as it was the only sensor, but the response seemed crisp (from memory) and when wired correctly heater noise is a non-event. Looking forward to running more of these in future!

Alan, without asking you to give away secrets, you often mention protection circuitry and so forth when reviewing other designs. This is very much after my own heart when it comes to PCB design. Can you list what the Spartan is and is not protected against? IE, what can it withstand, and what might damage it? /me goes to check the site. Nope, fair questions, not listed. Could you give the same specs for the PP2 unit pictured above? These specs would make for a nice comparison with Helmut's findings, IMO.

Fred.

For SLC PP2, these are just approximate numbers:
-I supply an external fuse, 2 separate power lines, eletronics get a 500ma fuse, sensor gets a 5 amp fuse
-TVS working voltage is ~20v, rated for 600W
-Power supply Reverse polarity protection diode is rated for 50A peak
-Inputs and outputs are all protected by current limiting resistors, 10ohms-5k ohms, some outputs like the pump cell drive to the sensor and virtual ground to the sensor need to be as low output resistance as possible for accuracy and it is at odds with proper protection so for that I use 10 ohms for those, usually I like to use 1k for the ordinary stuff.
-Inputs and outputs are protected with a ~175w TVS diode.

For Spartan Lambda Sensor, approximate numbers:
-I supply external fuse, 1 power line fused with 5amp fuse
-TVS working voltage ~20v, rated for ~350W
-Power supply Reverse polarity protection diode is rated for 30A peak
-Since the controller is soldered to the sensor, I just have current limiting resistors on the connections to the sensor, there is no TVS protection on those lines
-There are only 2 outputs on spartan lambda sensor, linear output and simulated narrowband output, both are current limited with a 5k resistor and a ~150W TVS diode.

For both units all inputs and outputs are protected with over current and over voltage protection, but to be honest I do no know just how good the protection is. There is an entire universe of study behind protection circuitry, I avoid opening that pandora's box by relying on parts that I have used in previous designs that gave me low instances of problems. I tend to like using discrete parts for voltage protection so I can physically see stuff blowing up and know more information about the failure; fuse gets blown, diode is burnt, etc...

[toalan]

Merry xmas to all

Alan:

That looks like a MUCH more professional and thought out PCB design than either the PLX or Innovate parts. Congtrats on out-hustling the big guys The layout style is actually similar to the LC-1.

I can't read the part numbers. But, right away it loos like there is a big Schottky diode and TVS on the input power rail. And, what looks like a smaller TVS array near the USB header ?

I didn't notice a ferrite bead near the USB. Are you using USB power as a signal only ?

I am split on the use of modern automotive grade tantalums vs aluminum caps.

On the spartan, you must get a lot of functionality out of that CPU. I assume it is one with the switched capacitor analog front-end ?


Nice work

Yes that is a TVS diode array for protecting the D+ and D- lines. Yes I am using USB power just to detect when the plug is inserted, using USB power just creates too much noise on the analog parts.

I have heard great things about the more recent high grade electrolytic caps, from a design point of view I have no problem using high grade electrolytics, but I fear that customers might mistaken them for low grade electrolytics and so I use tan caps just for image sake.

Spartan uses a plain vanilla Cortex M0, my SLC stuff uses PSOC that has the wacky switch capacitor stuff.

The PSOC design is is pretty much like almost any lambda controller design in terms of functional blocks; pump cell driven by DAC fed by PID, virtual ground is fed by a opamp biased at vdd/2 (in the SLC design it is a DAC but it never moves so I guess you can consider it an resistor divider buffered by an OPamp), pump current is fed through a instrumentation amp to the ADC. With PSOC I just build all the analog parts out of the switch capacitor blocks instead of having external discrete blocks. So the SLC design is a pretty plain vanilla design.

There is one unique feature of the SLC design that I have never talked about before and never seen implemented on any other controller. You can build an integrating ADC using the switch capacitor blocks, that in itself offers really awesome noise performance, but on the SLC design I set the heater PWM period to be a multiple of the ADC integration cycle so the heater switching noise is totally rejected. Whenever you switch the heater it creates a sh*t load of noise on the nernst cell and the pump cell since they are physically very close to the heater.. Also you can dither the PSOC DAC at a frequency multiple of the ADC integration cycle so you get extra DAC resolution for free. The noise performance of SLC is probably the best on the market, it is a really solid design for a noisy automotive environment, well atleast I think it is a great design but nobody has copied it yet so maybe I am wrong.

The Spartan design is a totally different approach, I am a bit of a prude to talk about it right now, but it carries over much of the noise performance of the SLC design and the idea of a single chip design.

[Fred]

Thanks for all of the info! Can the 5k resistors on the spartan outputs be lowered at all? Or is that value/size totally necessary? With that much output impedance, it's going to depend a lot on what the input circuit looks like, as to how sharp the response is/isn't. I'd feel most comfortable with them in the 500ohm region but if that doesn't result in a reliable setup, then it's obviously off the menu.

It was always a common procedure, in a certain other community, to use a 100ohm resistor inline with LC1 outputs to prevent issues, I'm assuming they're not very buffered? Or perhaps the "advice" was crooked advice?

[HelmutVonAutobahn]

Yep. I am seeing the newer aluminum "cans" on more and more OEM automotive applications, recently. And, not only in high ripple-current spots; where you might expect them.

Synchronizing the ADC with a big switching load is a normal thing in industrial designs. Sometimes, there is nothing else you can do.

IIRC, I remember some people getting better results with an LC-1 by installing a toriod choke in the heater ground (blue) line. And, a ~ 10ohm resistor in the signal ground line. There may have been a diode mixed int there. I might be inclined to toss a 2 or 3 ohm 10W resistor in there. Better to have a higher PWM duty cycle than a high peak current

[toalan]

Thanks for all of the info! Can the 5k resistors on the spartan outputs be lowered at all? Or is that value/size totally necessary? With that much output impedance, it's going to depend a lot on what the input circuit looks like, as to how sharp the response is/isn't. I'd feel most comfortable with them in the 500ohm region but if that doesn't result in a reliable setup, then it's obviously off the menu.

It was always a common procedure, in a certain other community, to use a 100ohm resistor inline with LC1 outputs to prevent issues, I'm assuming they're not very buffered? Or perhaps the "advice" was crooked advice?

Looking at the spartan PCB now, I am not sure what the actual output impedance is, the uC pin PWMs to an RC filter using 5k and 100nf, the output of the RC filter is TVS protected then goes through a 1k resistor to the outside. If My cap was significantly larger then the output impedance would be ~1k resistor, but since the capacitor is only 100nf, I am not sure. It is certainly lower than 6k (5k+1k) and certainly greater than 1k. I could not use a larger cap for the RC filter because I could only use an 0402 package due to space, and I wanted atleast X7R rated ceramics, the largest capacitance I could find was 100nf.

I have no idea about the LC1.

[toalan]

Synchronizing the ADC with a big switching load is a normal thing in industrial designs. Sometimes, there is nothing else you can do.

Dang you just sank my battleship, I thought my Synchronizing the ADC with the heater was original work.

[Fred]

ADC readings are synchronised with the engine in FreeEMS, mostly because of MAP, but it works for many of the inputs for various reasons in various cases. EG, starter current during cranking is cyclic with the compression of each cylinder, and heavily affects battery voltage, etc. I'm sure your work was original and self-discovered, though, as it was for me doing it in FreeEMS :-)

[DonTZ125]

Synchronizing the ADC with a big switching load is a normal thing in industrial designs. Sometimes, there is nothing else you can do.

Dang you just sank my battleship, I thought my Synchronizing the ADC with the heater was original work.

Well, you can certainly say your design follows "Industry Best Practices!"

[toalan]

Oh to be exact it is not synchronized, the ADC is an integrating ADC so the integrating cycle is set to to match the PWM frequency. This is something that is commonly used to measure the avg of household AC voltage, where the 1/integral-time is set to 50 or 60 hz.

I do not pick when and where to start the ADC conversion, the ADC conversion has it's own interrupt and is sampling all the time, when the heater PWM period matches the ADC integral time then the noise will be rejected.

I think the tech edge units actually synchronize the ADC to the switching heater switching, that is they do not run the ADC near when the heater switches, and only run the ADC after the heater load switches. This is not what I am doing.

[HelmutVonAutobahn]

There are lots of techniques.... interleaving, suppression, sequencing, etc.

It sounded like the one you were describing was using an integrating ADC with the load switching synchronized so that the effect of the transients is predictable, at minimum, or removed via symmetry, ideally.

An integrating delta-sigma ADC makes a LOT more sense in this application than the normal S/H + SAR. I have often thought that a delta-sigma ADC synchronized with the 4kHz Nernst cell measurement signal & a synched ~32kHz heater PWM drive would give the best feedback signal for a pump current control PID.

With the ADC running at 8kHz, the difference between successive sample gives the high/low Nernst resistance measurement. The average of every pair of successive samples gives the Nernst cell voltage.

i.e. everything "bad" SHOULD cancel out. .. on paper, at least