The BOM has this part number for the oscillator: SIT8918BE-13-33E-5.000000E
Most of this isn't too interesting, but to be pedantic in case I made an error reading the secret decoder ring:
Revision = "B"
Temperature range = "E" for Industrial
Output Drive Strength = "-" for default
Package size = "1" for 2.5x2.0 mm
Frequency stability = "3" for +/- 50 PPM
Supply Voltage = "33" for 3.3V
Feature pin = "E" for output enable
Frequency = 5.000000 MHz
Packing Method = "E" for 8 mm tape and reel
Looking at the datasheet https://www.sitime.com/products/datasheets/sit8918/SiT8918-datasheet.pdf it would seem that the output drive strength could be reduced significantly. If this part is the culprit, then perhaps that could help, and Parallax could make this change at some point if they think it might reduce their support costs down the line.
Jesse, please do look at lower frequencies if you have time, /3 and /2 submultiples especially.
The A&O app-note for the switching power supply attributes parasitic oscillation to a loop that involves the output impedance of the SMPS chip and the Schottky diode. It does not involve the main inductor or bypass capacitors, and indeed I found that adding more bypass to the FLIP SMPS did not help with the GPS effect. If the parasitic loop includes the diode it could be a potent generator of harmonics.of some fundamental frequency pumped by the switching transitions .
The note is applicable because the Schottky diode (Rohm RB068MM-40) is external to the control chip (AOZ1281DI), and it forms a potential resonant loop separate from the one that contains the main power inductor.
The bottom line is guidance about how to choose a C and an R for a snubber network across the diode. The app note says that the value of C should be about three times the intrinsic capacitance of the Schottky. From the attached graph for the RB068MM that is close to 100pF at 3.3V. So the snubber should use a 300pF capacitor. The second step requires knowing the frequency of the parasitic. From that, calculate the equivalent circuit inductance of the loop. I'll assume the 175MHz frequency. Maybe Jesse's SDR will locate a lower fundamental that might be better for this calculation.
Lp = 1 / ((2*pi*Fo)^2 * C) = 1/(6.28 * 175E6)^2 * 100E-12) = 8.3nH.
Then, choose a snubber resistor with ohms value equal to twice the inductive reactance at the parasitic oscillation frequency, so
Rs = 2 * (6.28 * 175E6 * 8.3E-9) = 18 ohms
So, the snubber across the diode would be 300pF in series with 18 ohms.
The theory behind it is a bit sketchy though. Has anyone else looked that over and have an opinion about it or a check on the calculations?
I hooked up a FLIP again under the pickup loop, but this time with the switching supply hard at work to drive a 4Ω load from its 3.3V protected output. As a result, the signal picked up by the loop above the SMPS section is much stronger and more consistent than it was when it only had to supply 15mA or so to run the flashing LEDs program.
There is a strong 75MHz component in the ringing, as seen in the 'scope capture. The 21st harmonic of that would fall at 1575MHz. The 7th harmonic at 525MHz. Both 525 and 1575 appeared in the SDR plots. A 7th harmonic is peculiar, but makes me wonder more about what might show up in the SDR plots. In particular, is there a strong component at 75MHz?
Was expecting to see some big gun at 75 but alas, I didn't see anything below 262 where I stopped the other night, all the way down to 1MHz. I did, however, find peaks every 5MHz (ending in 0 or 5) above 1575.42 (to the tuner's limit 1766). They weren't particularly strong, varying between -39 to -52 but definitely noticeable. Not sure if these are real or what??
Was expecting to see some big gun at 75 but alas, I didn't see anything below 262 where I stopped the other night, all the way down to 1MHz. I did, however, find peaks every 5MHz (ending in 0 or 5) above 1575.42 (to the tuner's limit 1766). They weren't particularly strong, varying between -39 to -52 but definitely noticeable. Not sure if these are real or what??
... does seem to point to the Osc module, - can you remove the Vcc and run the Prop on RC osc, just to see difference ?
(keep SMPS operating)
Can try this, but it may have to wait until I can bring it into work on Monday - not sure how to effectively do this without removing the osc from the board temporarily with a heat gun, which I don't have (alternative non-destructive suggestions welcome )
Use rcfast or tie reset pin to gnd. That will disable the osc. Varying loads on 3.3v header could simulate propeller current.
No, as I understand it, the external chip is an oscillator, not a crystal, with its own supply voltage. Perhaps the 3V3 to the oscillator power trace can be cut, and resoldered afterwards - I haven't chrcked the pcb layout.
Thanks for looking at that avsa242. I'm puzzled too that nothing showed up at the lower frequencies.
I went ahead and pulled the SIT8918 oscillator off the board, and ran a program at RCFAST and RCSLOW. The GPS is running off a 5V lab supply via a shielded cable, and the FLIP is running off USB power from the computer. The FLIP is located 2 inches from the GPS off to one side, and is not moved between program tests.
-- RCSLOW the GPS always keeps its lock.
-- RCFAST, the GPS generally keeps its lock.
I say generally because sometimes with RCFAST I would see the GPS lose lock for a while and then pick up again.
So, the SIT8918 is implicated as the culprit or at least an accomplice.
CON
'_clkmode = RCSLOW
_clkmode = RCFAST
PUB main
' alternate flashing p26 and p27 leds at 1 second interval
dira[27..26] := %11
outa[27..26] := %00
ctra := (%101<<26) + (26<<9) + 27
frqa := posx/clkfreq
'generate highest possible frequency on p0,p1 differential for clkfreq test
dira[0..1] := %11
ctrb := (%101<<26)+1
frqb := negx
repeat
So, the SIT8918 is implicated as the culprit or at least an accomplice.
Seems that way.
I wonder if other brands or part codes are better at 'less effect on GPS systems' ?
How well decoupled is the SIT8918 ?
Would a metal shield help here ?
I can well understand Parallax's dilemma that resulted in their choice of a timing reference for this board. There just aren't any tiny 5 MHz crystals about. My solution when I designed the Propeller Backpack was to use a small 10 MHz crystal, even though the Prop's PLLs are not characterized for an input frequency that high. Nonetheless, during the life of that product, I know of no instances where that caused a problem. I'd probably do it that way again, too, given restrictive size constraints.
I can well understand Parallax's dilemma that resulted in their choice of a timing reference for this board. There just aren't any tiny 5 MHz crystals about. My solution when I designed the Propeller Backpack was to use a small 10 MHz crystal, even though the Prop's PLLs are not characterized for an input frequency that high. Nonetheless, during the life of that product, I know of no instances where that caused a problem. I'd probably do it that way again, too, given restrictive size constraints.
-Phil
For the same reasons, my P8XBlade2 uses a 12MHz SMT xtal for 96MHz operation. Again, no reported problems, and I test using my Propeller OS.
I can well understand Parallax's dilemma that resulted in their choice of a timing reference for this board. There just aren't any tiny 5 MHz crystals about.
True, but there are resonators in (4.50mm x 2.00mm) packages, and at sub 20c/3k
Not as precise as a crystal, or as precise as those SIT8918 seem to be, but maybe ok for FLiP.
Of course, my personal preference is always for more precision, so I'd target some precision EPSON GPS oscillator module (62c/1k).
Paired with maybe a 74HCT393BQ, or I just notice Nexperia have a new series 74AHC1G4212, in a nifty SOT353 5 lead gull wing, but sadly they only show part codes for /10,/12,/14 taps... Seems a 74AHC1G4201 or 74AHC1G4202 would be ideal here....
There is a 74HC6323A, but only in SO8
or, ~32c can buy you a 2% 5MHz RC Osc, in a 3mm package, (with a free MCU)
My solution when I designed the Propeller Backpack was to use a small 10 MHz crystal, even though the Prop's PLLs are not characterized for an input frequency that high. Nonetheless, during the life of that product, I know of no instances where that caused a problem. I'd probably do it that way again, too, given restrictive size constraints.
For the same reasons, my P8XBlade2 uses a 12MHz SMT xtal for 96MHz operation. Again, no reported problems, and I test using my Propeller OS.
Parallax cannot really use something way outside spec, but they do have some choices here
** They can improve the spec, so that over a narrower Temp / Vcc span, other operating points are legal. This is very common in MCUs
** They could bin-select P1's and provide a faster version, that basically removes the slowest corner cases.
Addit: The 74HC6323A is readily available, but is a larger package and higher cost. Looks a bit EOL....
I've asked Nexperia about a 74AHC1G4202(?)
NJR have some Osc+Dividers, but do not seem well stocked.
Torex may be better, I see they have a [sample] button on the part code XC2164C51VMR-G, which I make as f/4, Fundamental, 20pF/20pF, SOT26 3,000: $0.281 "Information: When you request free samples, we send up to 50 of each part number. If you wish more than 50, please use online shopping."
Pair the XC2164C51VMR with a compact 20 or 25MHz xtal, (15~20c), or for better precision, an EPSON TG2016SBN Oscillator, 20 or 25MHz for ~62.35c, and it looks good...
...I just notice Nexperia have a new series 74AHC1G4212, in a nifty SOT353 5 lead gull wing, but sadly they only show part codes for /10,/12,/14 taps... Seems a 74AHC1G4201 or 74AHC1G4202 would be ideal here....
Addit:
I've asked Nexperia about a 74AHC1G4202(?)
Nexperia have replied : Yes we can offer /2 and /4 options for this part.
So that is positive news, this family are new parts, and low cost, and well suited to scaling GPS type Oscillators for MCU use.
74AHC1G42xx parts that exist are all priced at 6000 $0.1461, so they are cheaper than XC2164C51VMR-G, but the 74AHC1G42xx family has no Rf or Ci, Co.
A low cost, modest precision pairing would be Resonator (2.50mm x 2.00mm Murata, Abracon, ECS), Caps included, 18~19c, + external Rf.
EPSON TG2016SBN Oscillator would need external Rf and one Cs, for significant improvement in precision for ~ 43c more.
As I mentioned earlier the SIT8918B can be configured for slower edge rates. It's running WAY faster than needed for a measly 5 MHz reference clock. I don't know if it's really the culprit. If Parallax really cared about this a lab tech could probably narrow it down in an afternoon be doing some changes to the switcher and the reference clock.
As I mentioned earlier the SIT8918B can be configured for slower edge rates. It's running WAY faster than needed for a measly 5 MHz reference clock. I don't know if it's really the culprit.
I doubt the edges are the issue, as that would have a quite different spectrum effect.
If Parallax really cared about this a lab tech could probably narrow it down in an afternoon be doing some changes to the switcher and the reference clock.
Parallax should care about this, as it means one product they sell (FLiP) is not really compatible with another product they sell (GPS modules).
There are plenty of useful educational examples around FLiP and GPS.
As best I can tell, the Epson SG-210STF 5.0000ML, is pin compatible with SIT8918BE-13-33E-5.000000E, so it should allow a quick retrofit and re-check.
SG-210STF Price is broadly similar to SIT8918B (and strangely, more expensive than the more precise & smaller, clipped sine EPSON TG2016SBN - shows what volumes can do).
>I doubt the edges are the issue, as that would have a quite different spectrum effect.
If the experiment is easy to do then I typically like to try it. Just to explore the thought process - one potential concern is a 5 MHz clock generating harmonics at 1575.42 MHz GPS L1. This would be the 315th harmonic which is odd. I figured that sharper edges would generate more odd harmonics.
You may have other concerns e.g. what's inside of the oscillator, so replacing it with a different part entirely is something good to try as well.
As I mentioned earlier the SIT8918B can be configured for slower edge rates. It's running WAY faster than needed for a measly 5 MHz reference clock.
Agreed! The Sitime data sheet for the SIT8918BE-13-33E-5.000000E is quite explicit about the potential harmonic EMI effects. Also there is an app-note, AN10022-rise-and-fall-time.pdf. The default rise time into 5pF or 15pF is 1ns or less. (The Prop X1 input capacitance is ~6pF.)
To deal with it, SITIME offers versions of the part with lower drive strength, and the rise time can also be limited by means of an external capacitor. SITIME offers free samples, SIT8918BEL13-33E-5.000000E, so I ordered some. I want my flip back, and to satisfy curiousity. The L in the part number calls out the slowest rise time, so, according to the data sheet, that would take it down to ~4ns into 6pF.
I still don't want to say that the SITIME alone is the culprit. While it is more tolerant than before, now that the STIME oscillator is off the board, I can still find positions near the GPS where it loses or has trouble acquiring lock.
Just received the samples of the STTIME clock chip with the slower rise time option. Not today, but I'll try to coax one of these back onto the FLIP this weekend.
The sample package traveled far:
Sepang, Malaysia -> Shenzhen, China -> Incheon, Korea (customs clearance) -> Anchorage, Alaska -> Ontario, California -> San Pablo, California -> Berkeley, California.
Just received the samples of the STTIME clock chip with the slower rise time option. Not today, but I'll try to coax one of these back onto the FLIP this weekend.
The sample package traveled far:
Sepang, Malaysia -> Shenzhen, China -> Incheon, Korea (customs clearance) -> Anchorage, Alaska -> Ontario, California -> San Pablo, California -> Berkeley, California.
Quite the journey Look forward to the difference.
It's not easy to be 100% certain which oscillators do not have 100's MHz VCOs hidden inside them, but I can find this stocked item (Digiekey,Chipstop), I think is a Crystal-based one & should drop-in ? ( Even this may divide, but I think only from ~20MHz, based on Icc groupings in the data)
SG-210STF 5.0000ML EPSON OSC XO 5.000MHZ CMOS SMD 166 - Immediate $1.44
XO (Standard) 5MHz Standby (Power Down) CMOS 1.6 V ~ 3.6 V ±50ppm -40°C ~ 85°C 1.8mA (2.50mm x 2.00mm)
Not stocked, but with lower (±10ppm, ±15ppm) are these parts ECS-2520S33-050-EN ECS-2520S33-050-FN - looks like Parallax could request a sample.
If the spectrum problem persists with lower drive, the EPSON SG-210STF 5.0000 could be worth qualifying ?
Be nice if some simple retrofit can improve GPS interoperability.
There's a coincidence @jmg. That Epson part was originally specified and used on the early FLiP samples (got one here- it's identifiable as gold instead of black if anyone else has one), but fell foul of stocking issues for production deadlines.
Apparently efforts were ongoing to get supply sorted. I could check the latest status next week.
There's a coincidence @jmg. That Epson part was originally specified and used on the early FLiP samples (got one here- it's identifiable as gold instead of black if anyone else has one), but fell foul of stocking issues for production deadlines.
Apparently efforts were ongoing to get supply sorted. I could check the latest status next week.
Interesting - it might be easy to do some A-B comparisons around these GPS effects, if there are boards already fitted.
Interesting pictures. There's quite a difference. STTIME has a strong peak about 17dB above ambient at 1575MHz, whereas the same peak for the EPSON is only about 2dB above ambient. Of interest too is that odd harmonics of 5MHz are quite strong in the STTIME. Its even harmonics are there but less prominent, ~5 dB above ambient. On the other hand, it is the EPSON's even harmonics that are strong, about 10dB above ambient, but only about +2dB for its odd harmonics. Thus EPSON has strong peaks at 1570 and 1580, but largely misses GPS L1 at 1575.
The STTIME claims a rise/fall time of less than 1ns into 15pf at 3.3V, while the EPSON claims 3ns under the same conditions. There must be more to it than rise time alone to shape the harmonic content. Power supply is 3.3V*4.5mA maximum for the STTIME vs 3.3V*1.8mA maximum for the EPSON.
Oh my! stocking issues are a total bugaboo when you're ready for a production run. The distributers don't even list the slow-edge version of the STTIME, so it would have to be factory stock only.
Power supply is 3.3V*4.5mA maximum for the STTIME vs 3.3V*1.8mA maximum for the EPSON
That caught my eye too. Depending on any difference in underlying operating freq. of the two oscillators (jmg suggested 20MHz I think for the Epson), and of course the "actual" drive current; such difference at high speed would lead to rather different current fluctuations across the somewhat limited power plane.
Some adjustment to the osc local caps may be in order.
Comments
Most of this isn't too interesting, but to be pedantic in case I made an error reading the secret decoder ring:
Revision = "B"
Temperature range = "E" for Industrial
Output Drive Strength = "-" for default
Package size = "1" for 2.5x2.0 mm
Frequency stability = "3" for +/- 50 PPM
Supply Voltage = "33" for 3.3V
Feature pin = "E" for output enable
Frequency = 5.000000 MHz
Packing Method = "E" for 8 mm tape and reel
Looking at the datasheet https://www.sitime.com/products/datasheets/sit8918/SiT8918-datasheet.pdf it would seem that the output drive strength could be reduced significantly. If this part is the culprit, then perhaps that could help, and Parallax could make this change at some point if they think it might reduce their support costs down the line.
The A&O app-note for the switching power supply attributes parasitic oscillation to a loop that involves the output impedance of the SMPS chip and the Schottky diode. It does not involve the main inductor or bypass capacitors, and indeed I found that adding more bypass to the FLIP SMPS did not help with the GPS effect. If the parasitic loop includes the diode it could be a potent generator of harmonics.of some fundamental frequency pumped by the switching transitions .
http://www.aosmd.com/res/application_notes/power-ics/PIC-005.pdf
The note is applicable because the Schottky diode (Rohm RB068MM-40) is external to the control chip (AOZ1281DI), and it forms a potential resonant loop separate from the one that contains the main power inductor.
The bottom line is guidance about how to choose a C and an R for a snubber network across the diode. The app note says that the value of C should be about three times the intrinsic capacitance of the Schottky. From the attached graph for the RB068MM that is close to 100pF at 3.3V. So the snubber should use a 300pF capacitor. The second step requires knowing the frequency of the parasitic. From that, calculate the equivalent circuit inductance of the loop. I'll assume the 175MHz frequency. Maybe Jesse's SDR will locate a lower fundamental that might be better for this calculation.
Lp = 1 / ((2*pi*Fo)^2 * C) = 1/(6.28 * 175E6)^2 * 100E-12) = 8.3nH.
Then, choose a snubber resistor with ohms value equal to twice the inductive reactance at the parasitic oscillation frequency, so
Rs = 2 * (6.28 * 175E6 * 8.3E-9) = 18 ohms
So, the snubber across the diode would be 300pF in series with 18 ohms.
The theory behind it is a bit sketchy though. Has anyone else looked that over and have an opinion about it or a check on the calculations?
There is a strong 75MHz component in the ringing, as seen in the 'scope capture. The 21st harmonic of that would fall at 1575MHz. The 7th harmonic at 525MHz. Both 525 and 1575 appeared in the SDR plots. A 7th harmonic is peculiar, but makes me wonder more about what might show up in the SDR plots. In particular, is there a strong component at 75MHz?
(keep SMPS operating)
Use rcfast or tie reset pin to gnd. That will disable the osc. Varying loads on 3.3v header could simulate propeller current.
I went ahead and pulled the SIT8918 oscillator off the board, and ran a program at RCFAST and RCSLOW. The GPS is running off a 5V lab supply via a shielded cable, and the FLIP is running off USB power from the computer. The FLIP is located 2 inches from the GPS off to one side, and is not moved between program tests.
-- RCSLOW the GPS always keeps its lock.
-- RCFAST, the GPS generally keeps its lock.
I say generally because sometimes with RCFAST I would see the GPS lose lock for a while and then pick up again.
So, the SIT8918 is implicated as the culprit or at least an accomplice.
CON '_clkmode = RCSLOW _clkmode = RCFAST PUB main ' alternate flashing p26 and p27 leds at 1 second interval dira[27..26] := %11 outa[27..26] := %00 ctra := (%101<<26) + (26<<9) + 27 frqa := posx/clkfreq 'generate highest possible frequency on p0,p1 differential for clkfreq test dira[0..1] := %11 ctrb := (%101<<26)+1 frqb := negx repeatI wonder if other brands or part codes are better at 'less effect on GPS systems' ?
How well decoupled is the SIT8918 ?
Would a metal shield help here ?
-Phil
Not as precise as a crystal, or as precise as those SIT8918 seem to be, but maybe ok for FLiP.
Of course, my personal preference is always for more precision, so I'd target some precision EPSON GPS oscillator module (62c/1k).
Paired with maybe a 74HCT393BQ, or I just notice Nexperia have a new series 74AHC1G4212, in a nifty SOT353 5 lead gull wing, but sadly they only show part codes for /10,/12,/14 taps... Seems a 74AHC1G4201 or 74AHC1G4202 would be ideal here....
There is a 74HC6323A, but only in SO8
or, ~32c can buy you a 2% 5MHz RC Osc, in a 3mm package, (with a free MCU)
Parallax cannot really use something way outside spec, but they do have some choices here
** They can improve the spec, so that over a narrower Temp / Vcc span, other operating points are legal. This is very common in MCUs
** They could bin-select P1's and provide a faster version, that basically removes the slowest corner cases.
Addit: The 74HC6323A is readily available, but is a larger package and higher cost. Looks a bit EOL....
I've asked Nexperia about a 74AHC1G4202(?)
NJR have some Osc+Dividers, but do not seem well stocked.
Torex may be better, I see they have a [sample] button on the part code XC2164C51VMR-G, which I make as f/4, Fundamental, 20pF/20pF, SOT26 3,000: $0.281
"Information: When you request free samples, we send up to 50 of each part number. If you wish more than 50, please use online shopping."
Pair the XC2164C51VMR with a compact 20 or 25MHz xtal, (15~20c), or for better precision, an EPSON TG2016SBN Oscillator, 20 or 25MHz for ~62.35c, and it looks good...
So that is positive news, this family are new parts, and low cost, and well suited to scaling GPS type Oscillators for MCU use.
74AHC1G42xx parts that exist are all priced at 6000 $0.1461, so they are cheaper than XC2164C51VMR-G, but the 74AHC1G42xx family has no Rf or Ci, Co.
A low cost, modest precision pairing would be Resonator (2.50mm x 2.00mm Murata, Abracon, ECS), Caps included, 18~19c, + external Rf.
EPSON TG2016SBN Oscillator would need external Rf and one Cs, for significant improvement in precision for ~ 43c more.
Parallax should care about this, as it means one product they sell (FLiP) is not really compatible with another product they sell (GPS modules).
There are plenty of useful educational examples around FLiP and GPS.
As best I can tell, the Epson SG-210STF 5.0000ML, is pin compatible with SIT8918BE-13-33E-5.000000E, so it should allow a quick retrofit and re-check.
SG-210STF Price is broadly similar to SIT8918B (and strangely, more expensive than the more precise & smaller, clipped sine EPSON TG2016SBN - shows what volumes can do).
If the experiment is easy to do then I typically like to try it. Just to explore the thought process - one potential concern is a 5 MHz clock generating harmonics at 1575.42 MHz GPS L1. This would be the 315th harmonic which is odd. I figured that sharper edges would generate more odd harmonics.
You may have other concerns e.g. what's inside of the oscillator, so replacing it with a different part entirely is something good to try as well.
Agreed! The Sitime data sheet for the SIT8918BE-13-33E-5.000000E is quite explicit about the potential harmonic EMI effects. Also there is an app-note,
AN10022-rise-and-fall-time.pdf. The default rise time into 5pF or 15pF is 1ns or less. (The Prop X1 input capacitance is ~6pF.)
To deal with it, SITIME offers versions of the part with lower drive strength, and the rise time can also be limited by means of an external capacitor. SITIME offers free samples, SIT8918BEL13-33E-5.000000E, so I ordered some. I want my flip back, and to satisfy curiousity. The L in the part number calls out the slowest rise time, so, according to the data sheet, that would take it down to ~4ns into 6pF.
I still don't want to say that the SITIME alone is the culprit. While it is more tolerant than before, now that the STIME oscillator is off the board, I can still find positions near the GPS where it loses or has trouble acquiring lock.
The sample package traveled far:
Sepang, Malaysia -> Shenzhen, China -> Incheon, Korea (customs clearance) -> Anchorage, Alaska -> Ontario, California -> San Pablo, California -> Berkeley, California.
Quite the journey
It's not easy to be 100% certain which oscillators do not have 100's MHz VCOs hidden inside them, but I can find this stocked item (Digiekey,Chipstop), I think is a Crystal-based one & should drop-in ? ( Even this may divide, but I think only from ~20MHz, based on Icc groupings in the data)
SG-210STF 5.0000ML EPSON OSC XO 5.000MHZ CMOS SMD 166 - Immediate $1.44
XO (Standard) 5MHz Standby (Power Down) CMOS 1.6 V ~ 3.6 V ±50ppm -40°C ~ 85°C 1.8mA (2.50mm x 2.00mm)
Not stocked, but with lower (±10ppm, ±15ppm) are these parts ECS-2520S33-050-EN ECS-2520S33-050-FN - looks like Parallax could request a sample.
If the spectrum problem persists with lower drive, the EPSON SG-210STF 5.0000 could be worth qualifying ?
Be nice if some simple retrofit can improve GPS interoperability.
Apparently efforts were ongoing to get supply sorted. I could check the latest status next week.
Interesting - it might be easy to do some A-B comparisons around these GPS effects, if there are boards already fitted.
Noise floor,
datecode 1652 (FLiP pre-production with EPSON),
datecode 1712 (FLiP rev A with SITIME)
Will be interesting how Tracy Allen's sample SiTime works out.
The STTIME claims a rise/fall time of less than 1ns into 15pf at 3.3V, while the EPSON claims 3ns under the same conditions. There must be more to it than rise time alone to shape the harmonic content. Power supply is 3.3V*4.5mA maximum for the STTIME vs 3.3V*1.8mA maximum for the EPSON.
Oh my! stocking issues are a total bugaboo when you're ready for a production run. The distributers don't even list the slow-edge version of the STTIME, so it would have to be factory stock only.
That caught my eye too. Depending on any difference in underlying operating freq. of the two oscillators (jmg suggested 20MHz I think for the Epson), and of course the "actual" drive current; such difference at high speed would lead to rather different current fluctuations across the somewhat limited power plane.
Some adjustment to the osc local caps may be in order.
Hi Don,
That was a Siglent SSA3021X spectrum analyzer.