-
Hello, Today I swept all my attenuators on the spectrum analyzer using the tracking generator. I decided to share typical results for the various types so you can see typical response. The Crystek and TPI units were bought from Digikey and everything else was bought on Ebay. I tested other values of attenuators as well, 3dB and 6dB. The shapes are very similar with only minor variations. The worst of the bunch is the TTI units, which are fine up to about 350MHz, so I use them on my Rigol 100Mhz Scope only.
-
Thanks! But those ripples are there probably due to insufficient calibration of network analyzer, probably port match. You can try to calibrate (normalize in this case) system with 3..6 dB attenuators at tracking gen. output and SA input (at the measurement plane).
-
I lack a spectrum analyser and tracking generator and network analyser for doing this so I use my best fast reference level pulse generator and oscilloscope to test attenuators in the time domain by measuring their impulse response.
-
I decided to share typical results for the various types so you can see typical response. The Crystek and TPI units were bought from Digikey and everything else was bought on Ebay.
this is the first report i saw recommending ebay rather than digikey.I lack a spectrum analyser and tracking generator and network analyzer for doing this so I use my best fast reference level pulse generator and oscilloscope to test attenuators in the time domain by measuring their impulse response.
i think you'll need to terminate your attenuator with 50 ohm terminator to rule out reflection effect if using high impedance input scope. i have a theory thats saying we can do this using near perfect square wave as you suggested and then download the sampled data and then processed in FFT. first direct connection without a DUT, we get a FFT, thats our normalization factor or a reference. and then we compute FFT with the DUT in place, we divide both FFT's magnitude hence we get attenuation for various spectrum. the longer the FFT bin, we should get smoother and detailed respond. we also should be able to get phase respond from this. i tried this few years ago to check filter (or any DUT) respond using a FG, DSO and PC processing, but i get unclean result. maybe my FFT was not long enough, maybe my setup was not good enough and maybe my square wave was not square enough...
-
Thank you for taking the trouble to post these. I just wish Dave would setup a wiki where such things were easy to find.
-
Thank you for your tests. Perhaps add in the price of the attenuators so we get an idea of how they compare for the price.
I just did some tests myself. Note that I used my best cable but it still introduces error, especially above 3 GHz.
Mini Circuits VAT series of SMA 0-6 GHz attenuators - retail price of 11.95 USD each in small quantities.
Scale is 0.5 dB per division, 10,20,30 dB models measured.
10 dB
20 dB
30 dB
Mini Circuits UNAT+ series N 0-6 GHz attenuators - retail price of 15.95 USD each in small quantities.
Scale is 0.5 dB per division, 20 and 30 dB models measured.
20 dB
30 dB
Pasternack PE7025-30 BNC 30 dB attenuator - only rated to 2 GHz but tested to 3 GHz. Stupid money.
*Scale is 0.1 dB per division* (these work extremely well, especially for a BNC connector)
And for completeness some classic HP 8491A N attenuators, rated to 12.4 GHz and obviously stupid money.
*Scale 0.1 dB per division*
Measured 3,6 and 10 dB
3 dB
6 dB
10 dB
-
Thank you for your tests. Perhaps add in the price of the attenuators so we get an idea of how they compare for the price.
Mini Circuits VAT series of SMA 0-6 GHz attenuators - retail price of 11.95 USD each in small quantities.
Scale is 0.5 dB per division, 10,20,30 dB models measured.
The Crystek BNC units were $28, the TTI units were $9, The Ebay SMA units were about $8 each, the Ebay N units were $9-15 each, the Ebay BNC units were $12 each, The 5W unit was $20, the 10W unit was $23 and the 50W unit was $40. Ebay prices will vary by vendor of course.
In my working life we avoided buying attenuators over 20db because we often found the 30dB and greater units to have worse frequency response. This may not be true for all manufactures but it was our policy. We also noted that while MiniCircuits attenuators were generally good, they use very small resistors and therefore do not tolerate large pulses as well as some other manufactures. We avoided using them in pulsed applications but were great in CW or small signal work. Where I am living now, I cannot easily buy from US companies with DigiKey being an exception because of their free international shipping policy. Of course the absolute best attenuators are made by Barth. But they are very expensive, have excellent frequency response and are very robust.
https://www.ebay.com/itm/5-000-Volt-Pulse-Power-Attenuator-50pS-risetime-10x-20-dB-50-Ohms/192453839754?hash=item2ccf248b8a:g:n08AAOSw14xWQVVz
-
I also checked my Bird 8322 200 watt 30 dB attenuator. It is rated to 500 MHz(within half a dB) tested to 2 GHz, I see it meets spec to 600 MHz.
I was able to buy the Mini Circuit attenuators I have locally for a few dollars each, they seem pretty fair for the price I guess.
-
I also checked my Bird 8322 200 watt 30 dB attenuator. It is rated to 500 MHz(within half a dB) tested to 2 GHz, I see it meets spec to 600 MHz.
I was able to buy the Mini Circuit attenuators I have locally for a few dollars each, they seem pretty fair for the price I guess.
Can't beat them at that price. -
I also checked my Bird 8322 200 watt 30 dB attenuator. It is rated to 500 MHz(within half a dB) tested to 2 GHz, I see it meets spec to 600 MHz.
What VNA are you using?
I was able to buy the Mini Circuit attenuators I have locally for a few dollars each, they seem pretty fair for the price I guess. -
I also checked my Bird 8322 200 watt 30 dB attenuator. It is rated to 500 MHz(within half a dB) tested to 2 GHz, I see it meets spec to 600 MHz.
What VNA are you using?
I was able to buy the Mini Circuit attenuators I have locally for a few dollars each, they seem pretty fair for the price I guess.
Keysight FieldFox. -
In my working life we avoided buying attenuators over 20db because we often found the 30dB and greater units to have worse frequency response. This may not be true for all manufactures but it was our policy.
If the attenuator is built as a single stage that is certainly true unless special transmission line construction is used for the various elements which invariably means a completely custom design for them as well.
I am not sure why multiple stages could not be used. The tricks I have used in the past to improve the high frequency performance of terminations is to either add an attenuator which is better than the termination or use a long length of lossy coaxial able like RG-174; if the cable is long enough and the frequencies of interest are high enough, then no termination is even needed.
-
I try and be patient and only buy >15GHz rated SMA attenuators and these can be found on ebay (used) quite cheaply if you can be patient and spot a bargain. A high frequency attenuator like this probably won't be very accurate (could be out by 0.3dB for example) but it will give a flat response. Usually better than +/-0.05dB over 6GHz for a 20dB attenuator for example. I rarely pay more than £15 for a used SMA attenuator like this on ebay.
-
Quote
And for completeness some classic HP 8491A N attenuators, rated to 12.4 GHz and obviously stupid money.
These should be really good up to 6GHz if they are healthy. I'd expect to see no wobbly ripple, just a smooth/gradual slope of maybe up to 0.1dB across LF through to 6GHz.
So I think they should be better (smoother?) than the response in your plots. Maybe it is your cables? -
I try and be patient and only buy >15GHz rated SMA attenuators and these can be found on ebay (used) quite cheaply if you can be patient and spot a bargain. A high frequency attenuator like this probably won't be very accurate (could be out by 0.3dB for example) but it will give a flat response. Usually better than +/-0.05dB over 6GHz for a 20dB attenuator for example.
I have a pile of ancient N attenuators which have stamped calibration tags on them showing attenuation to 0.1 or 0.01 dB at different frequencies. It is very rare to need absolute tolerance as long as you know exactly what the attenuation is.
-
Quote
And for completeness some classic HP 8491A N attenuators, rated to 12.4 GHz and obviously stupid money.
These should be really good up to 6GHz if they are healthy. I'd expect to see no wobbly ripple, just a smooth/gradual slope of maybe up to 0.1dB across LF through to 6GHz.
So I think they should be better (smoother?) than the response in your plots. Maybe it is your cables?
As I mentioned in my post it is the cables - anything over 3 GHz and they suffer greatly. -
I recently bought a SMA gage set, and have put it to use on a Crystek CATTEN-06R0 6 dB SMA attenuator (from DigiKey). The female end is out of spec with the female dielectric protruding by 0.005". The gage set's instructions suggest a limit of 0.002" protrusion. The pin depths were OK.
My Mini-Circuits VAT-10+ meets the pin and dielectric depth specifications.
This is only a sample size of one for each manufacturer, so I don't know if I can draw any further conclusions. However, I do prefer the machining of the Mini-Circuits attenuator. -
Just a comment regarding BNC attenuators at high frequencies -
The scope input has 1MOhm || some Picofarads, at a PCB inside the unit.
Then, we have a piece of 50 line (from the socket at the scope to the input circuit at a PCB), ca 2cm
Then, we have the external BNC attenuator - another ca. 2 cm from match point to attenuator inside
So, we have about 4cm of a 50 Ohm line after the attenuator inside to the 1MOhm scope input on a PCB.
Lets assume a velocity factor of 0.6 - so this piece has an electrical length of ca. 6.5cm, this is a quarter
wavelength at around 1.2GHz.
Result: Using a BNC 50 Ohm thru terminator for anything even close to a GHz is asking for trouble, regardless
if the attenuator inside is properly built or not. I would say 500MHz is the absolute limit for something like this,
if you would like to believe whats on your screen. In riseterm terms, this is about 700ps.
If your signals are faster, get a better scope that has a 50Ohm input.
-
This thread spurred me to check mine.
Specs for the Tek 1:1 and 10:1 attenuators in the Tek Attenuator pdf above.
A Pintek PL-50N 1:1 bought recently is the middle one and rated to 2 GHz.
N to BNC adapters and a 500mm BNC cable used for tests.
SSA3032X set to sweep to 2.1 GHz to mirror the OP's tests.
Normalized reference screenshot.
Tek 1:1
Pintek PL-50N 1:1
Tek 10:1
-
The scope input has 1MOhm || some Picofarads, at a PCB inside the unit.
Then, we have a piece of 50 line (from the socket at the scope to the input circuit at a PCB), ca 2cm
Then, we have the external BNC attenuator - another ca. 2 cm from match point to attenuator inside.
So, we have about 4cm of a 50 Ohm line after the attenuator inside to the 1MOhm scope input on a PCB.
I have not examined any oscilloscopes which used a separate 50 ohm lead between the BNC and high impedance attenuator. They mostly either had a short flying lead which was usually part of a very small low value carbon composition resistor or the BNC was directly attached to the special printed circuit board with the high impedance attenuator.QuoteLets assume a velocity factor of 0.6 - so this piece has an electrical length of ca. 6.5cm, this is a quarter wavelength at around 1.2GHz. Result: Using a BNC 50 Ohm thru terminator for anything even close to a GHz is asking for trouble, regardless if the attenuator inside is properly built or not. I would say 500MHz is the absolute limit for something like this, if you would like to believe whats on your screen. In riseterm terms, this is about 700ps.
If your signals are faster, get a better scope that has a 50Ohm input.
The input capacitance of the high impedance buffer limits bandwidth up to about 500 MHz maximum. Oscilloscopes which are faster than this when the internal 50 ohm termination is used have a different arrangement where the input is switched by a microwave relay between the low impedance and high impedance circuit paths which duplicates how the old Tektronix 485 works; it is 250MHz in 1 megohm mode and 350MHz in 50 ohm mode and it has no internal feedthrough termination.
So the difference between using a good external feedthrough termination and the internal termination should be small. I cannot see any difference on my fastest high impedance oscilloscope but it is only 300MHz.This thread spurred me to check mine.
Specs for the Tek 1:1 and 10:1 attenuators in the Tek Attenuator pdf above.
A Pintek PL-50N 1:1 bought recently is the middle one and rated to 2 GHz.
I am surprised that Tektronix 10x (20dB) attenuator was that good but maybe I should not have been. Tektronix specified it as <1.1 VSWR at 1GHz and <1.2 VSWR at 2GHz. That feedthrough was specified at <1.1 VSWR at 250MHz and <1.2 VSWR at 500MHz. I have a pair of their lower frequency (<1.1 VSWR at 100MHz) 5W feedthroughs which were actually manufactured by someone else; ProbeMaster used to sell the exact same ones.
Their GR attenuators were specified up to 1GHz, their N attenuators up to 12GHz, and their SMA attenuators up to 18GHz.
-
The Tek 10x is rated to 2 GHz like the 1:1 Pintek that I've found cleaner for square wave overshoot than the Tek 1:1.This thread spurred me to check mine.
Specs for the Tek 1:1 and 10:1 attenuators in the Tek Attenuator pdf above.
A Pintek PL-50N 1:1 bought recently is the middle one and rated to 2 GHz.
I am surprised that Tektronix 10x (20dB) attenuator was that good but maybe I should not have been. Tektronix specified it as <1.1 VSWR at 1GHz and <1.2 VSWR at 2GHz. That feedthrough was specified at <1.1 VSWR at 250MHz and <1.2 VSWR at 500MHz. I have a pair of their lower frequency (<1.1 VSWR at 100MHz) 5W feedthroughs which were actually manufactured by someone else; ProbeMaster used to sell the exact same ones.
Their GR attenuators were specified up to 1GHz, their N attenuators up to 12GHz, and their SMA attenuators up to 18GHz.
I might do those sweeps again as at 5dB/div it's hiding the full story.
Need to show under 1dB/div like The Steve did.
-
The scope input has 1MOhm || some Picofarads, at a PCB inside the unit.
Then, we have a piece of 50 line (from the socket at the scope to the input circuit at a PCB), ca 2cm
Then, we have the external BNC attenuator - another ca. 2 cm from match point to attenuator inside.
So, we have about 4cm of a 50 Ohm line after the attenuator inside to the 1MOhm scope input on a PCB.
I have not examined any oscilloscopes which used a separate 50 ohm lead between the BNC and high impedance attenuator. They mostly either had a short flying lead which was usually part of a very small low value carbon composition resistor or the BNC was directly attached to the special printed circuit board with the high impedance attenuator.QuoteLets assume a velocity factor of 0.6 - so this piece has an electrical length of ca. 6.5cm, this is a quarter wavelength at around 1.2GHz. Result: Using a BNC 50 Ohm thru terminator for anything even close to a GHz is asking for trouble, regardless if the attenuator inside is properly built or not. I would say 500MHz is the absolute limit for something like this, if you would like to believe whats on your screen. In riseterm terms, this is about 700ps.
If your signals are faster, get a better scope that has a 50Ohm input.
The input capacitance of the high impedance buffer limits bandwidth up to about 500 MHz maximum. Oscilloscopes which are faster than this when the internal 50 ohm termination is used have a different arrangement where the input is switched by a microwave relay between the low impedance and high impedance circuit paths which duplicates how the old Tektronix 485 works; it is 250MHz in 1 megohm mode and 350MHz in 50 ohm mode and it has no internal feedthrough termination.
So the difference between using a good external feedthrough termination and the internal termination should be small. I cannot see any difference on my fastest high impedance oscilloscope but it is only 300MHz.
300MHz can be still OK. On a faster scope (Tek MSO4104, Keysight DSOS604, ...) differences are striking from a few 100MHz up.
This thread spurred me to check mine.
Specs for the Tek 1:1 and 10:1 attenuators in the Tek Attenuator pdf above.
A Pintek PL-50N 1:1 bought recently is the middle one and rated to 2 GHz.
I am surprised that Tektronix 10x (20dB) attenuator was that good but maybe I should not have been. Tektronix specified it as <1.1 VSWR at 1GHz and <1.2 VSWR at 2GHz. That feedthrough was specified at <1.1 VSWR at 250MHz and <1.2 VSWR at 500MHz. I have a pair of their lower frequency (<1.1 VSWR at 100MHz) 5W feedthroughs which were actually manufactured by someone else; ProbeMaster used to sell the exact same ones.
Their GR attenuators were specified up to 1GHz, their N attenuators up to 12GHz, and their SMA attenuators up to 18GHz.
Attenuators and terminators are completely different issues; Most attenuators go up to about 2GHz even when they are low quality. Shunting a coax by a 50 Ohm resistor, with the line reaching a few centimeters more and them terminated by 1MOhm and 10pF is a completely different story. -
Attenuators and terminators are completely different issues; Most attenuators go up to about 2GHz even when they are low quality. Shunting a coax by a 50 Ohm resistor, with the line reaching a few centimeters more and them terminated by 1MOhm and 10pF is a completely different story.
I agree but it is not a dire problem when the high impedance buffer will limit bandwidth to abut 500 MHz anyway.
High impedance oscilloscope inputs and probes are actually designed to compensate for the impedance discontinuity and capacitance assuming that either an external 25 ohm source is used or an external 50 ohm source and internal 50 ohm termination is used. You should get practically identical results when using the recommended high impedance passive probes and 25 ohm source and a 50 ohm source with a feedthrough attenuator or internal termination although this gets increasingly chancy above 200 to 300 MHz.
This is why Tektronix had several probe series with letter revisions. The A and B versions were specifically compensated for later oscilloscopes even when they had the same bandwidth as the earlier ones.