原创 Correcting a power-hungry USB

The company I had been working for then had designed and created a USB-powered data-acquisition product. The product worked great, but when it came time to verify its ability to meet the current-draw USB requirement for suspend mode, it failed miserably. Although the calculated current draw was well below the 500µA current spec limit for low-power suspend mode, the actual draw was about 5mA. The design engineer had spent weeks measuring things and replacing parts but nothing worked.

After a few frustrating weeks the engineer was told to enlist my help, so, together we made yet more measurements. A DMM was used to verify all the usual voltages, and a scope to ensure something wasn't oscillating, unfortunately, everything looked fine. One day, while bending over the suspended unit I noticed that the current dropped from about 5mA to 1mA. Stand up straight and the current returned to 5mA; was this some kind of oscillation and our body capacitance affecting the circuit's Q? Most analogue guys have experienced something like this at one time or another. More probing with the scope and lightly running our fingers over PCB pads and components revealed nothing was oscillating.

At that time I remembered having read that glass encapsulated diodes can act as photodiodes. This was an SMT board with no glass diodes, but we did have an LED on the front panel. Cover the LED and the current dropped, uncover it and it rose again, voila!

This LED had a grounded cathode and was connected to the bus driver through a resistor. As it turned out, the LED would generate enough photoelectric voltage to drive the tri-stated output into a linear mode such that the bus driver IC's internal circuitry would draw 4mA off of the 5V supply. The tri-stated output was still a very high impedance and so it would not sink the LED's photoelectric-generated current, so the LED's voltage was truly a function of the light entering its lens.

The output impedance of the LED's photoelectric source must have been very high, attempts to measure it with a scope showed nothing, the 10Meg input impedance must have loaded the source until there was an immeasurable voltage left. Tried again with a bargain basement, hand-held DMM with a 10Meg input impedance and it indicated only a few millivolts, the DMM still loading the source. When the LED output was measured with a higher quality 6½ digit DMM with GOhms of input impedance a whopping 1.4V was indicated.

I had remembered reading an application note by TI about something similar to this, "Implications of Slow or Floating CMOS Inputs" (SCBA004C, February 1998).

This note included a graph indicating that the IC's supply is a function of the logic input voltage level; similar graphs are included in the datasheets of other logic devices. But, ours was not an input but a tri-stated output, apparently a tri-stated output that is driven by some external voltage can still affect the quiescent current of the device. Who knew?

The solution was simple, place a 22k resistor from the LED's anode to ground. This gave the light-induced LED current a place to go without biasing the tri-stated output into conduction.

The data acquisition card now behaved well, even in an environmental chamber operating from -40°C to +60°C.

Glenn Fasnacht has worked with analogue, mixed signal, switched mode power supply design for test and measurement products since 1975. Since 1998, he has worked stints in project management at Reliance Electric, Smith Meters, Akron Standard, Keithley Instruments, IOTech, and VTI Instruments. He has four US patents.

He submitted this article as part of Frankenstein's Fix, a design contest hosted by EE Times (US).