According to the datasheet, the LM4562 has a GBW of 55MHz.
I simulated a test circuit in LTSpice (see attached).
The simulation indicates I should get a 3dB bandwidth of around 5.7MHz.

When I test the circuit for real, I measure the 3dB bandwidth to be only around 400kHz.

I can't figure out why. Any suggestions?

Note that to get the simulation to work, I had to specify a small series resistance for the supply voltages (0.1 Ohm).
Without that, I get a 'Floating Nodes' error. I assume this is down to some issue with the LTSpice model for the LM4562.

 

I can not see your picture well. Attach the file so we can try it.
It looks like the gain is 10/ stage. "GBW of 55MHz." 55mhz for gain =1, 5.5mhz for gain=10, 550k for gain=100. See Figure 106.
With two stages: (-3db at 5.5mhz) followed by (-3db at 5.5mhz) = (-6db at 5.5mhz)
The amplifier has a "volts/uS" rating on the output. How fast the output can move. So at low frequency it can output a large signal but at about 8mhz it can only make 1vp-p. See Fig 105. At 10Vp-p it can only make 800khz.
There can be other problems. These to are first thing.

 

Here are the files...

 

You will also need LM4562.lib

 

The LTSpice simulation result appears to be about what I would expect from the info in the datasheet.

The device is not driving a heavy load - the test circuit is terminated with 100K.
The input test signal amplitude is small - around 20mV rms.
I've tried different supply voltages +/- 5V and +/- 12V.
I've checked the output signal using a 'scope and can see no sign of unwanted oscillation, clipping etc.
The device supply pins are properly decoupled with 0.1uF caps using short leads.
I've tried two different samples of the device.

 

The measured mid-band gain looks correct.
I've checked using both a scope and a wide band AC millivoltmeter.

 

What is the bandwidth of just the first stage in the real circuit?

 

LTspice calculated everything correctly. The bandwidth of one operational amplifier at a level of 3 dB is equal to the frequency of a unit gain of an opamp divided by the gain. If you use two identical stages of operational amplifiers, then each of the operating ones will have only 1.5 dB. Those. less bandwidth in two stages.
You have done a small signal analysis. At the large output signal at the output of the second opamp, the band will already be due to the limitation of another parameter - the slew rate of the output voltage.

 

LTspice calculated everything correctly. The bandwidth of one operational amplifier at a level of 3 dB is equal to the frequency of a unit gain of an opamp divided by the gain. If you use two identical stages of operational amplifiers, then each of the operating ones will have only 1.5 dB. Those. less bandwidth in two stages.
You have done a small signal analysis. At the large output signal at the output of the second opamp, the band will already be due to the limitation of another parameter - the slew rate of the output voltage.

In my real test circuit, the signal at the output of the second opamp is less than 1V rms. I could see no visible evidence of waveform distortion due to slew rate limiting.
Does the LTSpice model not take the effect of slew rate into account?

I'm starting to wonder if the devices I am using are fakes. They came from an eBay supplier.
I'm not sure how to identify a fake from the markings. I've attached a close up photo.

 

I've tried reducing the input signal level by 20dB. That gives an output level of less than 100mV.
The measured 3dB bandwidth does not change.
If the device cannot handle small signals such as this without slew rate limiting then what's the point of specifying a gain-bandwidth product of 55MHz in the data sheet?
It looks to me as if the devices I am using have a GBW of something more like 5MHz.

 

I constructed the dependence of the output amplitude divided by the input amplitude on the input amplitude.

 

I constructed the dependence of the output amplitude divided by the input amplitude on the input amplitude.

Thanks for that. It's a useful transient analysis.

The application is an amplifier forming part of the front end of an AC millivoltmeter.
The amp is preceeded by a FET impedance converter and a set of switchable attenuators.
The maximum input signal level at the amp input will never exceed 1mV rms.
I would be happy if I can achieve a 3dB bandwidth of 1MHz.

I adjusted the parameters of the LTSpice transient analysis accordingly:-
A 1MHz sine wave with an amplitude ranging from 1mV peak to 10mV peak.
As far as I can tell, the results indicate no sign of slew rate limiting.

This is in agreement with my observations using a scope on the real test circuit:-
My signal generator only goes up to 1MHz.
At this frequency, I only see signs of slew rate limiting when the signal on the output exceeds around 1v rms.
This corresponds to an input level of around 20mV.
This is 20 times higher than the amp will need to handle.

If I reduce the input level to 2mV rms, the measured 3dB bandwidth still only 400kHz.

The results make me think that slew rate is not the limiting factor.
Unless it turns out that the devices are fakes, I'm still baffled.

 

In my real test circuit, the signal at the output of the second opamp is less than 1V rms. I could see no visible evidence of waveform distortion due to slew rate limiting.
Does the LTSpice model not take the effect of slew rate into account?

I'm starting to wonder if the devices I am using are fakes. They came from an eBay supplier.
I'm not sure how to identify a fake from the markings. I've attached a close up photo.

Im guessing your dealing with fake parts or seconds that didn’t make the cut during production testing. This is not unusual when purchasing from non reputable dealers.

Purchase instead from digikey, mouser, etc.

I once was a member of a monthly subscription electronic kit service (hackerboxes) that resold knock off parts or parts that wouldn’t pass testing. A massive amount of time was spent just trying to get things to work just to find solder balls, solder bridges, fake parts etc.

The concept of the company was cool but their execution was poor due to part issues. I stopped my subscription after 2 months. It just wasn’t worth my time or money to futz with fake parts.

YMMV

 

If the operational amplifier can be modeled as a single pole amplifier (has a single time constant response) than the gain-bandwidth product is equal to the unity gain-frequency. If that is the case the LM4562 would have a unity gain at 55MHz, a bandwith of 5.5Mhz at a gain of 10, a bandwidth of 550khz at a gain of 100, etc.

 

An update:-

I ordered some more LM4562. This time from Mouser.
Mouser are an approved distributor for Texas Instruments who took over National Semiconductor in 2011.

After replacing the suspect device, the test circuit now performs as expected.
The measured 3dB bandwidth matches the LTspice simulation

For me, this is conclusive proof that the devices bought from eBay are counterfeit.

Three years ago I built a 'State Variable' ultra low distortion oscillator.
I used LM4562 op amps bought from the same eBay supplier.
I tested those devices yesterday and guess what? I found the same bandwidth problem!
The original oscillator design had a max frequency of 150kHz but I couldn't get it to work properly at high frequencies.
The two integrator op amps didn't appear to be operating as expected. I ended up having to limit the frequency to 100kHz.
At the time I didn't suspect I had been sold fake op amps.

In future I will only buy components from trusted suppliers.

 

Three years ago I built a 'State Variable' ultra low distortion oscillator.
I used LM4562 op amps bought from the same eBay supplier.
I tested those devices yesterday and guess what? I found the same bandwidth problem!

In future I will only buy components from trusted suppliers.

ebay is pro-buyer. I'd file a claim with ebay, and explain the situation, you'll likely get your money back. I'd also leave negative feedback and low-star ratings with the seller. The sale of counterfeit parts won't stop if these companies aren't held accountable.

BTW - good work designing your circuit to use the parts you need, but not going overboard. You're likely saving cost by doing so. Also good detective work figuring out the problem!