Op amp noise revisited - the nuts and bolts
In the last column (Simulating op amp noise) I showed how to do simple and fast simulations on op amp circuits to get a handle on what kind of noise performance to expect. As can be expected with any simulation the results are only as good as:
- The models used
- How accurate the representation of the circuit that is being modeled really is
A question arose about the model input current noise generator shown in that article and whether it should properly be modeled as an independent current source for the plus and negative inputs. An excellent question.
Way back in the last century, IC designers cleverly figured out that they could compensate for the bias current of their bipolar input op amps by injecting an equal and opposite current into the inputs, thus canceling the input current that the op amp user's circuit sees (Figure 1). This was a huge step forward – all of a sudden our op amp circuits got much more stable and forgiving. As the decades passed, this basic circuit was improved upon for temperature stability and repeatability. The earliest implementations added a fair amount of input noise ; the latest generation has really improved that a lot.
Figure 1: A very rough sketch of how the basic input bias current cancellation scheme works. Input transistors Q2 and Q3 are part of the amplifier's input stage. These transistors have a finite base current. By adding a suitably biased transistor Q1, a small but equal bias current can be injected that very closely matches the base currents of Q2 and Q3. This basic scheme has been used and improved upon for decades now.
The basic current compensation circuit idea is shown in very simple form in Figure 1. The op amp input transistors Q2 and Q3, drawn here as NPNs have some base current, which in fact is quite large on low-noise op amps, because to get the input noise voltage down, the collector current of the input transistors must be large. Q1 (with suitable biasing) is designed to inject an equal and opposite current into the op amp's input, thus canceling out the base current of Q2 and Q3.
On the data sheet you will see input bias current numbers being given a range of plus and minus numbers with the typical being zero. This compensation circuit is designed to exactly cancel the typical base current of the input transistors, but due to circuit tolerances has some range that can be expected in any individual op amp sample.
The noise of these compensation circuits has been improved so much so that the majority of the input current noise now can come from Q1, the input bias current cancellation circuit.
Regarding the test circuit of the previous article in this series (Reference , figure 1), which is the same basic configuration that TI's Art Kay used in his noise simulation series:
Mr. Kay states: "Strictly speaking, there are two current noise sources. The degree however, to which these sources are correlated is not always clear from the product data sheet." 
We get some more information on this from the LT1028 data sheet where it states:
"The cancellation circuitry injects two correlated current noise components into the two inputs. With matched source resistors the injected current noise creates a common-mode voltage noise and gets rejected by the amplifier. With source resistance in one input only, the cancellation noise is added to the amplifier’s inherent noise." .
The LT1028 data sheet further gives curves for total noise with matched input resistances and without. The Analog Devices AD8675 op amp that I also simulated does not go into this detail but it seems highly likely that it has much the same characteristics – there is probably a high degree of correlation in the input noise current due to the input bias current compensation circuits.
Reflection on Simulations Past
What if the op amps didn't have correlated input current, what would be the result?
Adding two current sources to the AD8675 op amp circuit of the previous article results in the circuit of Figure 2.
Figure 2: A proposed circuit that adds totally uncorrelated input current noise to the AD8675 simulation of the last article (In11 and In12 shown above).
Simulating Figure 2 and comparing to the previous result results in the curve in Figure 3. The results are certainly within the accuracy of the models used and are nearly identical.
Figure 3: Results of two simulations, the first with a single uncorrelated input current noise generator and the second with two uncorrelated current noise sources in both the Plus and Minus op amp Inputs. The results are nearly identical for the circuit configuration of figure 2.
This is because the source resistances of the proposed application circuit are not balanced from the +IN and -IN terminals of the op amps and only the 3500-Ω source resistance side dominates. This imbalance was the result of a conscious decision to limit the voltage noise addition of the gain-setting resistors, so that only the 3500-Ω source resistance would dominate the noise. Also, since the AD8675 op amp is operating well below its Ropt point , its voltage noise is dominating - not its current noise.
So this particular circuit configuration showed that the LT1028 would not be the optimum amplifier for this circuit because its current noise is dominating the total noise since it would be operating well above its Ropt point. The AD8675 is a better choice here because of the opposite – its current noise is so low that it has nearly no effect in this application circuit.