Cpk Setting Tolerance - Cart before the Horse? (Wifi Routers)

Hami812

Engineering Program Mg
I'm struggling with a Colleague approach to something related to testing Wifi Routers who wishs to perform a Gage R&R on some Routers that have Wifi. There are numerous testers that need to be "Qualified" for production and a gage RnR is required. My colleague suggests, in order to set the "limits for pass and fail" that We should be taking a set of brand new routers which have nothing wrong with them, and testing each one 1 time. THere are 25-30 routers x 1 test (all known good routers). THe parametric I am concerned is the Wifi Power level measured in Dbm value from -10 to -75. UCL is fixed at -10, but there is no LCL. He wants to adjust the LCL (Mfg says it should be -55 dbm, but they are resistant to using this!) tolerance using Cpk by adjusting the Cpk value for each tester instead of trying to get all the testers to reproduce the same result, or ver similar results, then later doing Cpk. Each antena for 5G for Transit (TX) and Receive (RX) and 2.4G Wifi reads a different value for RX and TX. THere are 6 Antennas (3 for each Technology type 2.4 and 5 g) all internally located inside. I perform these tests and what I am finding is inconsistencies between the testers. For example Tester A tests 5 db higher than tester B, but only for Antena 2 (example below). IN most cases the testers are off by at least 5-8 db for both TX Antenna 1,2,3 when compared, and RX 1,2,3 for both 2.4 and 5 G Wifi specifications, sometimes more. This is a ton if your know RF. a movement of just 3 db is double the power or halving it depending on what direction it takes. What is the proper way to conduct preparation for a customer to do a Gage Study. Our customer will do Fixed runs and Rotational runs and they are saying There are 20% more false fails than is expected. What should We do first? I dont think setting tolerances should be done until process is more stable. Can someone weigh in and help me with a methodology for preparing for a Gage R&R? DO you need more info? Uploaded example file

Example data looks like this. Excel formaulas are used

RSSI_Value_5G_1_TX
RSSI_Value_5G_2_TX
RSSI_Value_5G_3_TX

dBm
dBm
dBm
STR
-10
-10
-10
UCL
-46
-62
-49
LCL
-43.0882
-53.2353
-46.8824
AVG
0.7535
2.1611
0.7693
STDEV
-42.0000
-50.0000
-45.0000
MAX
-45.0000
-58.0000
-48.0000
MIN
14.63790968
6.668803916
15.9812135
Cpku
1.288136051
1.351907188
0.917581636
Cpkl
1.288136051
1.351907188
0.917581636
Cpk
 

Attachments

  • Elsmar cove example.xls
    26.5 KB · Views: 160

Bev D

Heretical Statistician
Leader
Super Moderator
Lets start by clearly defining some terms and requirements for Gage R&R testing.

Specifications or tolerances come from the functioning of the product. What limits are required on the power for the successful use of the router in the intended use conditions? Specifications and tolerances do NOT come from the product performance capability...

You use the term LCL and UCL, by this do you mean upper and lower CONTROL limit? these limits are not "Set", they are determined by the actual performance. you measure many parts then statistically determine the limits of expected performance. In your case I would assume that an I, MR chart would be the best approach. However, control limits are NOT used in Gage R&R testing.

Remember that in Gage R&R testing you will be assessing the TESTER and not the Router. However, you need to have routers that span nearly the full range of variation in power to get useful meaningful results from your Gage R&R.

Since you are assessing the Tester's ability to measure the power of the routers, you will need multiple measurements of each router used in the test. The repeatability and reproducibility are assessed from the multiple measurements of each router. If you only have one measurement per router, you can only assessing the router.

If you measure each router only once with multiple testers, you will only get an assessment of the total R&R of all testers. You will not have an assessment of the repeatability of a tester which is of primary importance in the R&R study.

you might find this paper on Measurement Systems Analysis valuable.
 

Hami812

Engineering Program Mg
Lets start by clearly defining some terms and requirements for Gage R&R testing.

Specifications or tolerances come from the functioning of the product. What limits are required on the power for the successful use of the router in the intended use conditions? Specifications and tolerances do NOT come from the product performance capability...

You use the term LCL and UCL, by this do you mean upper and lower CONTROL limit? these limits are not "Set", they are determined by the actual performance. you measure many parts then statistically determine the limits of expected performance. In your case I would assume that an I, MR chart would be the best approach. However, control limits are NOT used in Gage R&R testing.

Remember that in Gage R&R testing you will be assessing the TESTER and not the Router. However, you need to have routers that span nearly the full range of variation in power to get useful meaningful results from your Gage R&R.

Since you are assessing the Tester's ability to measure the power of the routers, you will need multiple measurements of each router used in the test. The repeatability and reproducibility are assessed from the multiple measurements of each router. If you only have one measurement per router, you can only assessing the router.

If you measure each router only once with multiple testers, you will only get an assessment of the total R&R of all testers. You will not have an assessment of the repeatability of a tester which is of primary importance in the R&R study.

you might find this paper on Measurement Systems Analysis valuable.
Bev, this is extremely useful and has been partly the line of thinking i have, but been reluctant to pursue without some confidence. Youve helped me greatly yet again. I really...really appreciate this feedback.. i will provide more guidance to the approach to be taken soon... it makes sense what ylur saying and will read the above before replying with more questions as needed.
 

Hami812

Engineering Program Mg
Lets start by clearly defining some terms and requirements for Gage R&R testing.

Specifications or tolerances come from the functioning of the product. What limits are required on the power for the successful use of the router in the intended use conditions? Specifications and tolerances do NOT come from the product performance capability...

You use the term LCL and UCL, by this do you mean upper and lower CONTROL limit? these limits are not "Set", they are determined by the actual performance. you measure many parts then statistically determine the limits of expected performance. In your case I would assume that an I, MR chart would be the best approach. However, control limits are NOT used in Gage R&R testing.

Remember that in Gage R&R testing you will be assessing the TESTER and not the Router. However, you need to have routers that span nearly the full range of variation in power to get useful meaningful results from your Gage R&R.

Since you are assessing the Tester's ability to measure the power of the routers, you will need multiple measurements of each router used in the test. The repeatability and reproducibility are assessed from the multiple measurements of each router. If you only have one measurement per router, you can only assessing the router.

If you measure each router only once with multiple testers, you will only get an assessment of the total R&R of all testers. You will not have an assessment of the repeatability of a tester which is of primary importance in the R&R study.

you might find this paper on Measurement Systems Analysis valuable.
Bev,

The UCL and LCL is part of this CPk math as i recall where the limits are subtracted from. So this below shows the fail limit as -50 wheras the Mfg is -55dbm. So when these fails go back to Mfg, they will get kicked back along with a hefty fee to us. This is ludicrous why the factory spec is simply not nailed up and used instead of doing CPk on known good population to set the limits for the bad!! (goes back to what you were saying)

RSSI_Value_5G_1_TXRSSI_Value_5G_2_TXRSSI_Value_5G_3_TXUUT_SNdBmdBmdBmSTR-10-10-10UCL-48-60-50LCL-43.4828-49.8966-44.8966AVG1.12192.66381.5663STDEV-42.0000-45.0000-43.0000MAX-46.0000-55.0000-48.0000MIN9.9483840844.9923553927.426678221Cpku1.3421609831.2642697751.086114997Cpkl1.3421609831.2642697751.086114997Cpk -45-51-44-45-50-47-44-48-46-45-52-48-45-54-45-46-51-46-43-54-46-45-50-47-43-48-43-44-54-47-44-53-47-43-52-46-42-50-44-44-51-43-42-50-43-43-50-44-43-45-43-42-47-43-43-50-44-42-50-44-43-51-44-45-55-47-42-47-45-43-48-46-43-47-44-43-47-45-43-46-43-43-46-43-43-50-45





. What my colleague does instead of doing an X-Bar, MR chart to see the voice of the process using spc, they take say 15-20 modems with wifi, and set the CP AFTER running the tests and gathering the data for these 6 antennas (6 for Receive, and 6 for transmit) each modem. Then Stddev is done and 4 times Stddev is performed, but only at the lower end. If you recall the higher dbm value is nailed up at -10 dbm which, by law, these modems will never hit ( exceeds 1 watt at -30). So their goal is to get to 1.33 CPk which is nearly impossible to get because the modems being tested are all perfectly good. So what happens here is the threshold setting becomes much lower than say -55 (50 in the case above)which as i mentioned is the mfg spec limit and SHOULD BE USED. When they play with the CPk value game it never hits 1.33, but instead they make an incremental change, usually closer to 1.0 than 1.33. What occurs when this change is made is a yield of false failures that hits 35-40 percent. When retested these fallout to nearly zero. Doesnt this tell them they are taking the wrong approach to this? NOT! My colleague insists that setitng CPk is proper way to go. But will need to get them to see the Spc light... im working it. Any further thoughts Bev? Your awesome...a d thanks again!
 
Last edited:
Top Bottom