# Output Voltage Minimum Limit of LM358 as Error Amplifer

I am using one side of the LM358 as an error amplifier with a gain of 1, biased according to this MIT lesson for differential amplifiers.

The results below show me tuning the reference signal to see if I get the desired output error. My error = SENSE + OFFSET - REF. It appears to saturate at an output voltage (the error signal) of 500mV.

Why is this? I've checked the datasheet I have for the LM358 but cannot seem to make sense of the operating limitations, since this shows the output voltage conditions with a load attached (I am operating with no load):

What am I missing? Where on the datasheet can I find information about the minimum output voltage?

My supply is +5V to GND, so I expected the output to reach zero and stay there when my calculated error reached the negative zone.

• Look on a block diagram (usually in data sheet) at the output transistor arrangement inside the opamp. Pull down capability is limited. Try adding a load to ground. The LM358 is NOT a "rail to rail" amplifier even though it is at least notionally single supply. A whiff of negative supply relative to ground can help muchly. A Heath Robinson but sometimes useful 'fix' is a diode in the nehative power lead with ground at diode anode and Vdd of opamp at diode cathode - about 0.6V below ground. Place a decoupling cap across the diode. You now have Vcc ~~= 4.3V and Vee = ~~-0.6V wrt ground. – Russell McMahon Apr 1 '18 at 12:38
• @RussellMcMahon thank you for your input. Looking at the FBD it is not immediately apparent that the pull down capability is limited. I understand that it is not a rail-to-rail amplifier, but a minimum VOUT of 0.5 is still 10% of the supply, which I figured was far too much. Without having the capability to supply a negative VCC instead of ground, you are suggesting the diode solution, correct? (Note: there is no drain supply, Vdd, on the opamp, just Vcc) – YNGVV Apr 2 '18 at 5:44
• I believe you are attempting to mix 3 voltages but superposition does not apply to the + input , which sees two voltage R dividers that load each other while you could be saturating your output. WHere is U1A out data? Was it noise free? – Tony Stewart Sunnyskyguy EE75 Apr 3 '18 at 7:36
• Hi @TonyStewartEEsince1975, thank you for your input. I am only mixing two voltages at the + input : OFFSET (also written as BIAS) + SENSE, and this appears to work because my actual ERROR signal matches my calculated error. It also worked in simulation. As for the data for the U1A, that is my LM358. I did attach the datasheet and included a snippet of the output voltage limitations section, which is the area of interest. Yes, these signals were noise free, from what I could see on the scope. – YNGVV Apr 3 '18 at 7:49
• At low Voit and low Vdifferential at high Vdd values the LM358 will sink 12 uA worst case. Datasheets TI - Fairchild - OnSemi. Typical values are higher BUT Murphy doesn't care. With R4 = 10k, 12 uA = 0.12V. Exactly what you need depends on the inverting input voltage BUT at low Vout the LM358 clearly gets 'rather wimpy'. – Russell McMahon Apr 4 '18 at 11:18