Sensitivity of opamps to air coupled EM fields, especially of the LM4562/LME497XY family
This has been (and will be) a long story that started back in the year 2008 when I started with intensive testing of various opamps concerning their susceptibility to electromagnetic fields in the air or from the inside of the instruments. Quite a long row of opamps was tested, including uA741, uA1458, LM358, OP97, TL071, TL072, AD829, AD844, AD797, AD825, LT1028, LT1122, OPA134, OPA2134, OPA627, OPA211, OPA827, LM6171, LM4562, LME49710, LME49720, NE5532. So they were both BJT and JFET input stage opamps, slow and fast, low supply current or standard supply current. Some of them performed very well and were almost unaffected by a “normal” lab or home environment EM field that surrounded the test PCB or box, some were performing “moderately” and some were performing very poorly, reacting to a slightest change of EM field in their vicinity, like changing their position on a test bench or approaching a hand or just a move of the air or of the body at a 1m distance. In general, JFET input opamps have had no problems and the behavior of BJT input opamps was very different, from excellent immunity to horrible sensitivity to a slightest change in their vicinity. As a conclusion, absolutely worst behavior was that of the LM4562/LME49710/LME49720 family, bought several times in the past 12 years, directly from the manufacturers or authorized distributors. No fakes, no cheap buys. Every time I published some of the results, it started strong reactions and dissatisfaction, especially of the people involved in audio production who have been using those parts that did not perform well.
The parts were measured in the lab where I used to work 12 years ago and then in my home office lab. Various soundcards were used, various PC's both desktop and notebooks battery operated, as well as analog oscilloscopes, DSO stand-alone oscilloscopes and USB scopes. The results were consistent regardless instrumentation used. The best performers re EM immunity were all JFET opamps tested (TL07X, OPA134, OPA627, OPA827, AD825), from BJT opamps excellent were AD797, LT1028, AD844. LM6171 and NE5532 were acceptable, while LM4562 and LME497X0 failed completely.
OK, let's see the story.
1st round, March 2008
Back in 2008 I decided to test open-loop gain linearity according to the circuit published in Analog Devices datasheet of the OP177 precision opamp, page 9. The circuit is shown in Fig.1. It has signal gain of -1 but it releases feedback loop by increasing noise gain (1M/10R resistors in original schematics). I did not go that far with noise gain because I wanted to test opamps with lower OL gain as well, so I used ratio of the noise gain divider as 100k/10ohm. So my circuit has had signal gain of -1 and noise gain of 86dB. As the noise gain is increased, feedback action is suppressed (there is less feedback) and inherent distortion and other inherent linear and non-linear parameters of the opamp become more evident. Input signal gain is -1 (inverting), however the opamp inherent noise is amplified of 86dB, i.e. 20 000x, with respect to opamp frequency response at such high gain. Because soundcard input sensitivity was about 1V, I added the 25dB divider to the output of the test circuit to be able to utilize higher opamp's output voltage.
The 1st test circuit was built on a universal test board, without a ground plane. The board was placed above larger Aluminum sheet metal that was connected with a signal ground in 1 point and provided some electrostatic shielding. The measuring setup was single ended, so currents flowing through signal cables shield unfortunately create small voltage drop that is added to the signal voltage and displayed as lines at mains frequency and its multiples. However, these mains spuriae are constant and same for all parts under test, so the test conditions remain equal for all parts under test.
Fig.1. Opamp test circuit, signal gain -1, noise gain 86dB
LT1028
Fig.2. LT1028 with noise gain 86dB and signal gain -1
LT1028 was an excellent bipolar input opamp with very low noise and very low distortion. We can see some mains system spuriae, low distortion of 0.03% (please take into account that the equivalent gain is 86dB), one of the lowest noise from all parts under test and no unexpected spectral interference lines.
OPA627
Fig.3. OPA with noise gain 86dB and signal gain -1
OPA627 is a very expensive JFET input opamp with low noise and low ditortion. We can see noise level about 10dB higher than with LT1028, low distortion about 0.04% and no unexpected spectral interference lines.
LME49710
Fig.4. LME49710 with noise gain 86dB and signal gain -1
LME49710 from National Semiconductor was a new part at the time and it was introduced with very high expectations and accompanied by high appraisal among audiophiles. So I ordered samples and put them into the test. It was a big disappointment. As we can see from the spectral plot, the distortion is low, however there is a forest of symmetrically distributed spectral lines, every 10Hz. What is that? And why specifically this opamp, all samples?
Fig.5. LME49710 noise gain 86dB, zero line
The measurement was repeated with no input signal, but the forest of spectral lines placed in 10Hz multiples remained there.
Fig.6. AD797 noise gain 86dB, zero line
Another bipolar low noise opamp was tested to compare, the AD797 by @scott wurcer . Zero line is clean, with only some mains lines that are there from the method setup.
I tested much more opamps in the 1st round, but those most interesting I have shown here. I posted those results in 2008 and received doubts about the LME49710 results. Wrong samples, bad PCB bypassing, shielding, you name it. It was difficult for many to digest that LME49710 result was bad, even after the fact that the test conditions were same for all the opamps under test.
So the test was repeated and the test PCB was placed in a completely shielded small metal box. The level of spuriae decreased, however the differences between opamps remained similar, with LME49710 still having forest of spuriae, now mostly at every 100Hz multiples. Excellent result again was that of LT1028.
Fig.7. LT1028 noise gain 86dB, signal gain -1, metal shielded box
2nd round of measurements
was made in 2013, with new opamp samples and almost same results as in 2008. Different soundcard, different measuring place, nothing new, just confirmation of previous results.
3rd round of measurements, January 2016
another samples, now dual opamps, and new measurements, now not only spectrum analysis, but also time domain measurements. Spectra were again similar to those we have already seen, but I have added oscilloscope measurements and I think they were interesting.
NE5532
Fig.8. NE5532 noise gain 86dB signal gain -1, input shorted.
NE5532 was added to the test and we can see spectral lines at every 100Hz, but they are about 20dB lower than that of LM4562 in Fig.9.
LM4562
Fig.9. LM4562 noise gain 86dB signal gain -1, input shorted. High spuriae at every 100Hz and much smaller spuriae at every 10Hz.
LM4562 100Hz sine 8Vp-p
Fig.10. LM4562 86dB noise gain, signal gain -1, 100Hz sine. We can see that every 10ms we have a spike, something triggers the opamp and it looks like if it worked for a while like a comparator without feedback. These sharp and short spikes are responsible for the forest of harmonics at 100Hz multiples that we could see in Fig.9.
Fig.11. LM4562 time record now made by a PicoScope USB oscilloscope. Same spikes, every 10ms, are visible.
Fig.12. LM4562, even the Picoscope low resolution spectral analysis catches the spikes and repetitive spectral lines every 100Hz. Frustrating.
Fig.13. OPA2134 tested clean in the same setup.
Fig.14. Time record of the LM4562 zero line. We can see both spikes with repetition frequency 100Hz and also the wider one with 10Hz repetition.
Battery power
Readers have suggested that the spikes are triggered from the power supply residual ripple. So I supplied the test board from two 9V batteries and helas – the spectrum remained unchanged.
Fig.15. LM4562 battery power supply
OK, the battery supply did not help, so what about further suggestions? They suggested that the problem is in the poor universal PCB and poor supply bypassing.
So I designed, ordered and paid the double-sided PCB with metal through-holes and top shielding ground plane.
Fig.16. New test circuit that comprises 86dB noise gain/-1 signal gain circuit and also 40dB signal gain circuit, for dual opamps.
Fig.17. And this is the photo of the new test circuit
The test procedure was repeated, amplitude of multiples of 100Hz spuriae in LM4562 plots was reduced, but they were still there. I was again instructed that the problem is in my PCB design and in the socket that is used for the LM4562 …...... I better add no comments.
4th round of measurements, January 2020
OK, I am stupid and cannot design the PCB to make that poor LM4562 happy enough. So let's move forward, I ordered from experts and from the source, directly from TI, the LME49720NABD Evaluation Board.
Fig.18. TI Evaluation Board schematics (it can be found only in the archive LM4562 datasheet)
Fig.19. LME49720NABD Evaluation Board
Nice board, isn't it? Short traces, SMD bypass capacitors right at the opamp pins, this will be a breakthrough, right? (… it was not, as we shall see)
and I have also put my double-sided PCB test board, with the 86dB noise gain circuit, into the test and placed it in a metal sheet box to provide shielding.
Fig.20. My 86dB noise gain test PCB in a shielded box
Let's start with the “proper evaluation board” LME49720NABD from the TI manufacturer. As seen in Fig.18, it has gain of mere -1, 10k/10k inverter. So there should be no issues. Nice PCB, proper supply bypassing, what else and better could be done.
So this is the output from the board OUT_1 measured with input of the board shorted by a jumper JP2.
Fig.21. LME49720NABD output with input shorted.
Sometimes it looks like this,
however
Fig.22. LME49720NABD output with input shorted.
Sometimes it looks like this. Depends. Depends on a slightest move of the board on my test bench and also on my hand approaching the board. Translated – the opamp input is catching EM field in its vicinity and triggers on 50Hz mains frequency field zero crossings. Now it depends on EM field intensity near the opamp what happens. So it has nothing in common with supply bypassing, with PSU ripple (we have seen the same issue with battery power in Fig.15) and even placing this board into Cu shielding box did not help. The only cure would probably be the iron box to shield magnetic compound of the EM field that is triggering the LME49720NA like a comparator.
So, no happy fortune with the TI professional evaluation board. Let's go back to my 86dB noise gain double-sided test PCB, placed in a metal sheet box. The similar set of measurements as in the 1st test round, in 2008, was made. The opamps under test were TL072, OPA2134, NE5532 and LM4562. Spectral plots are properly calibrated in dBV, this time. Plot results are following, without further comments, except for the one – LM4562 was the only one to show 100Hz multiples spuriae.
Fig.23. Noise gain 86dB test, TL072 output with input shorted.
Fig.24. Noise gain 86dB test, OPA2134 output with input shorted.
Fig.25. Noise gain 86dB test, NE5532 output with input shorted.
Fig.26. Noise gain 86dB test, LM4562 output with input shorted.
Fig.27. Noise gain 86dB test, TL072 output for 100Hz input.
Fig.28. Noise gain 86dB test, OPA2134 output for 100Hz input.
Fig.29. Noise gain 86dB test, NE5532 output for 100Hz input.
Fig.30. Noise gain 86dB test, LM4562 output for 100Hz input.
Conclusion
Since 2008, I made 4 rounds of tests of opamps in a test circuit with -1 signal gain (inverter) but with 86dB noise gain, to test inherent linearity of opamps. The opamps tested were uA741, uA1458, LM358, OP97, TL071, TL072, AD829, AD844, AD797, AD825, LT1028, LT1122, OPA134, OPA2134, OPA627, OPA211, OPA827, LM6171, LM4562, LME49710, LME49720, NE5532. As a side effect it was found that some of the opamps, especially the LM4562/LME497X0 family were extremely sensitive to EM field of mains frequency in the vicinity of the test board. This was blamed on improper test board design, improper supply bypassing or PSU ripple. All these arguments were proven wrong by a newly designed double-sided PCB with ground plane, battery power instead of DC PSU and professional TI evaluation board purchase. None of this solved the issue of some of the opamps.
As a conclusion, there was never a problem with JFET input opamps, there was never a problem with bipolar input opamps AD797, LT1028, AD844, there were slight issues with LM6171 and NE5532 and there were big issues, always, with the LM4562/LME497X0 family. This family of opamps is extremely sensitive to EM fields in their vicinity, mains frequency 50Hz triggers them during zero crossing and they send narrow spikes to their output, in 100ms distances and of amplitude that depends on shielding. This can be cured extremely well shielded box that shields not only against electric field component, but also against magnetic field component. The problem may remain hidden, sometimes it disappears with respect to momentary EM field conditions, but can be seen also for gains as low as -1 in the professional TI LME49720NABD evaluation board. The problem was observed by various soundcards and PCs in 2 independent measuring places and also by using analog and DSO scopes. It looks like a duck, it quack like a duck, so it would be a duck.
Your comments are welcome, but please let me not to reply the questions that are already covered in this text, it was an exhausting job.
This has been (and will be) a long story that started back in the year 2008 when I started with intensive testing of various opamps concerning their susceptibility to electromagnetic fields in the air or from the inside of the instruments. Quite a long row of opamps was tested, including uA741, uA1458, LM358, OP97, TL071, TL072, AD829, AD844, AD797, AD825, LT1028, LT1122, OPA134, OPA2134, OPA627, OPA211, OPA827, LM6171, LM4562, LME49710, LME49720, NE5532. So they were both BJT and JFET input stage opamps, slow and fast, low supply current or standard supply current. Some of them performed very well and were almost unaffected by a “normal” lab or home environment EM field that surrounded the test PCB or box, some were performing “moderately” and some were performing very poorly, reacting to a slightest change of EM field in their vicinity, like changing their position on a test bench or approaching a hand or just a move of the air or of the body at a 1m distance. In general, JFET input opamps have had no problems and the behavior of BJT input opamps was very different, from excellent immunity to horrible sensitivity to a slightest change in their vicinity. As a conclusion, absolutely worst behavior was that of the LM4562/LME49710/LME49720 family, bought several times in the past 12 years, directly from the manufacturers or authorized distributors. No fakes, no cheap buys. Every time I published some of the results, it started strong reactions and dissatisfaction, especially of the people involved in audio production who have been using those parts that did not perform well.
The parts were measured in the lab where I used to work 12 years ago and then in my home office lab. Various soundcards were used, various PC's both desktop and notebooks battery operated, as well as analog oscilloscopes, DSO stand-alone oscilloscopes and USB scopes. The results were consistent regardless instrumentation used. The best performers re EM immunity were all JFET opamps tested (TL07X, OPA134, OPA627, OPA827, AD825), from BJT opamps excellent were AD797, LT1028, AD844. LM6171 and NE5532 were acceptable, while LM4562 and LME497X0 failed completely.
OK, let's see the story.
1st round, March 2008
Back in 2008 I decided to test open-loop gain linearity according to the circuit published in Analog Devices datasheet of the OP177 precision opamp, page 9. The circuit is shown in Fig.1. It has signal gain of -1 but it releases feedback loop by increasing noise gain (1M/10R resistors in original schematics). I did not go that far with noise gain because I wanted to test opamps with lower OL gain as well, so I used ratio of the noise gain divider as 100k/10ohm. So my circuit has had signal gain of -1 and noise gain of 86dB. As the noise gain is increased, feedback action is suppressed (there is less feedback) and inherent distortion and other inherent linear and non-linear parameters of the opamp become more evident. Input signal gain is -1 (inverting), however the opamp inherent noise is amplified of 86dB, i.e. 20 000x, with respect to opamp frequency response at such high gain. Because soundcard input sensitivity was about 1V, I added the 25dB divider to the output of the test circuit to be able to utilize higher opamp's output voltage.
The 1st test circuit was built on a universal test board, without a ground plane. The board was placed above larger Aluminum sheet metal that was connected with a signal ground in 1 point and provided some electrostatic shielding. The measuring setup was single ended, so currents flowing through signal cables shield unfortunately create small voltage drop that is added to the signal voltage and displayed as lines at mains frequency and its multiples. However, these mains spuriae are constant and same for all parts under test, so the test conditions remain equal for all parts under test.
Fig.1. Opamp test circuit, signal gain -1, noise gain 86dB
LT1028
Fig.2. LT1028 with noise gain 86dB and signal gain -1
LT1028 was an excellent bipolar input opamp with very low noise and very low distortion. We can see some mains system spuriae, low distortion of 0.03% (please take into account that the equivalent gain is 86dB), one of the lowest noise from all parts under test and no unexpected spectral interference lines.
OPA627
Fig.3. OPA with noise gain 86dB and signal gain -1
OPA627 is a very expensive JFET input opamp with low noise and low ditortion. We can see noise level about 10dB higher than with LT1028, low distortion about 0.04% and no unexpected spectral interference lines.
LME49710
Fig.4. LME49710 with noise gain 86dB and signal gain -1
LME49710 from National Semiconductor was a new part at the time and it was introduced with very high expectations and accompanied by high appraisal among audiophiles. So I ordered samples and put them into the test. It was a big disappointment. As we can see from the spectral plot, the distortion is low, however there is a forest of symmetrically distributed spectral lines, every 10Hz. What is that? And why specifically this opamp, all samples?
Fig.5. LME49710 noise gain 86dB, zero line
The measurement was repeated with no input signal, but the forest of spectral lines placed in 10Hz multiples remained there.
Fig.6. AD797 noise gain 86dB, zero line
Another bipolar low noise opamp was tested to compare, the AD797 by @scott wurcer . Zero line is clean, with only some mains lines that are there from the method setup.
I tested much more opamps in the 1st round, but those most interesting I have shown here. I posted those results in 2008 and received doubts about the LME49710 results. Wrong samples, bad PCB bypassing, shielding, you name it. It was difficult for many to digest that LME49710 result was bad, even after the fact that the test conditions were same for all the opamps under test.
So the test was repeated and the test PCB was placed in a completely shielded small metal box. The level of spuriae decreased, however the differences between opamps remained similar, with LME49710 still having forest of spuriae, now mostly at every 100Hz multiples. Excellent result again was that of LT1028.
Fig.7. LT1028 noise gain 86dB, signal gain -1, metal shielded box
2nd round of measurements
was made in 2013, with new opamp samples and almost same results as in 2008. Different soundcard, different measuring place, nothing new, just confirmation of previous results.
3rd round of measurements, January 2016
another samples, now dual opamps, and new measurements, now not only spectrum analysis, but also time domain measurements. Spectra were again similar to those we have already seen, but I have added oscilloscope measurements and I think they were interesting.
NE5532
Fig.8. NE5532 noise gain 86dB signal gain -1, input shorted.
NE5532 was added to the test and we can see spectral lines at every 100Hz, but they are about 20dB lower than that of LM4562 in Fig.9.
LM4562
Fig.9. LM4562 noise gain 86dB signal gain -1, input shorted. High spuriae at every 100Hz and much smaller spuriae at every 10Hz.
LM4562 100Hz sine 8Vp-p
Fig.10. LM4562 86dB noise gain, signal gain -1, 100Hz sine. We can see that every 10ms we have a spike, something triggers the opamp and it looks like if it worked for a while like a comparator without feedback. These sharp and short spikes are responsible for the forest of harmonics at 100Hz multiples that we could see in Fig.9.
Fig.11. LM4562 time record now made by a PicoScope USB oscilloscope. Same spikes, every 10ms, are visible.
Fig.12. LM4562, even the Picoscope low resolution spectral analysis catches the spikes and repetitive spectral lines every 100Hz. Frustrating.
Fig.13. OPA2134 tested clean in the same setup.
Fig.14. Time record of the LM4562 zero line. We can see both spikes with repetition frequency 100Hz and also the wider one with 10Hz repetition.
Battery power
Readers have suggested that the spikes are triggered from the power supply residual ripple. So I supplied the test board from two 9V batteries and helas – the spectrum remained unchanged.
Fig.15. LM4562 battery power supply
OK, the battery supply did not help, so what about further suggestions? They suggested that the problem is in the poor universal PCB and poor supply bypassing.
So I designed, ordered and paid the double-sided PCB with metal through-holes and top shielding ground plane.
Fig.16. New test circuit that comprises 86dB noise gain/-1 signal gain circuit and also 40dB signal gain circuit, for dual opamps.
Fig.17. And this is the photo of the new test circuit
The test procedure was repeated, amplitude of multiples of 100Hz spuriae in LM4562 plots was reduced, but they were still there. I was again instructed that the problem is in my PCB design and in the socket that is used for the LM4562 …...... I better add no comments.
4th round of measurements, January 2020
OK, I am stupid and cannot design the PCB to make that poor LM4562 happy enough. So let's move forward, I ordered from experts and from the source, directly from TI, the LME49720NABD Evaluation Board.
Fig.18. TI Evaluation Board schematics (it can be found only in the archive LM4562 datasheet)
Fig.19. LME49720NABD Evaluation Board
Nice board, isn't it? Short traces, SMD bypass capacitors right at the opamp pins, this will be a breakthrough, right? (… it was not, as we shall see)
and I have also put my double-sided PCB test board, with the 86dB noise gain circuit, into the test and placed it in a metal sheet box to provide shielding.
Fig.20. My 86dB noise gain test PCB in a shielded box
Let's start with the “proper evaluation board” LME49720NABD from the TI manufacturer. As seen in Fig.18, it has gain of mere -1, 10k/10k inverter. So there should be no issues. Nice PCB, proper supply bypassing, what else and better could be done.
So this is the output from the board OUT_1 measured with input of the board shorted by a jumper JP2.
Fig.21. LME49720NABD output with input shorted.
Sometimes it looks like this,
however
Fig.22. LME49720NABD output with input shorted.
Sometimes it looks like this. Depends. Depends on a slightest move of the board on my test bench and also on my hand approaching the board. Translated – the opamp input is catching EM field in its vicinity and triggers on 50Hz mains frequency field zero crossings. Now it depends on EM field intensity near the opamp what happens. So it has nothing in common with supply bypassing, with PSU ripple (we have seen the same issue with battery power in Fig.15) and even placing this board into Cu shielding box did not help. The only cure would probably be the iron box to shield magnetic compound of the EM field that is triggering the LME49720NA like a comparator.
So, no happy fortune with the TI professional evaluation board. Let's go back to my 86dB noise gain double-sided test PCB, placed in a metal sheet box. The similar set of measurements as in the 1st test round, in 2008, was made. The opamps under test were TL072, OPA2134, NE5532 and LM4562. Spectral plots are properly calibrated in dBV, this time. Plot results are following, without further comments, except for the one – LM4562 was the only one to show 100Hz multiples spuriae.
Fig.23. Noise gain 86dB test, TL072 output with input shorted.
Fig.24. Noise gain 86dB test, OPA2134 output with input shorted.
Fig.25. Noise gain 86dB test, NE5532 output with input shorted.
Fig.26. Noise gain 86dB test, LM4562 output with input shorted.
Fig.27. Noise gain 86dB test, TL072 output for 100Hz input.
Fig.28. Noise gain 86dB test, OPA2134 output for 100Hz input.
Fig.29. Noise gain 86dB test, NE5532 output for 100Hz input.
Fig.30. Noise gain 86dB test, LM4562 output for 100Hz input.
Conclusion
Since 2008, I made 4 rounds of tests of opamps in a test circuit with -1 signal gain (inverter) but with 86dB noise gain, to test inherent linearity of opamps. The opamps tested were uA741, uA1458, LM358, OP97, TL071, TL072, AD829, AD844, AD797, AD825, LT1028, LT1122, OPA134, OPA2134, OPA627, OPA211, OPA827, LM6171, LM4562, LME49710, LME49720, NE5532. As a side effect it was found that some of the opamps, especially the LM4562/LME497X0 family were extremely sensitive to EM field of mains frequency in the vicinity of the test board. This was blamed on improper test board design, improper supply bypassing or PSU ripple. All these arguments were proven wrong by a newly designed double-sided PCB with ground plane, battery power instead of DC PSU and professional TI evaluation board purchase. None of this solved the issue of some of the opamps.
As a conclusion, there was never a problem with JFET input opamps, there was never a problem with bipolar input opamps AD797, LT1028, AD844, there were slight issues with LM6171 and NE5532 and there were big issues, always, with the LM4562/LME497X0 family. This family of opamps is extremely sensitive to EM fields in their vicinity, mains frequency 50Hz triggers them during zero crossing and they send narrow spikes to their output, in 100ms distances and of amplitude that depends on shielding. This can be cured extremely well shielded box that shields not only against electric field component, but also against magnetic field component. The problem may remain hidden, sometimes it disappears with respect to momentary EM field conditions, but can be seen also for gains as low as -1 in the professional TI LME49720NABD evaluation board. The problem was observed by various soundcards and PCs in 2 independent measuring places and also by using analog and DSO scopes. It looks like a duck, it quack like a duck, so it would be a duck.
Your comments are welcome, but please let me not to reply the questions that are already covered in this text, it was an exhausting job.
Last edited:
