|
||||||||||||||||||||||||||||||||||||
|
|
||||||||||||||||||||||||||||||||||||
|
||||||||||||||||||||||||||||||||||||
|
|
Bonnie C. Baker, Senior Applications Engineer, Data Acquisition Products Division, Burr-Brown Corporation I remember the days when I could count on enough good analog circuit questions to create quite an afternoon of adventures in the lab. I would use theory to predict the analog outcome in terms of stability, gain or noise levels and try to match it with reality. The mixed-signal ADC and DAC questions were riddled with digital timing problems to conquer, but seldom the complex noise or stability calculations that analog circuits provided. Since the mixed-signal circuits still required an analog front end (gain and filtering), most calculations were still handled in the analog domain. The entrance of the 16-bit converter offered somewhat of a front-end noise reduction challenge, but it did not rival the battle being waged in the pure analog domain. Has analog been replaced with sophisticated mixed-signal devices? Then a new player joined the team. This player was literally thrown over the wall, as it were (for those who aren't IEEE fans). From the IC designer to the poor unsuspecting IC user. It came in the form of a one-bit digitizer that would output a 24-bit word. I once asked an audience of engineers, in a technical seminar, what 224 was equal to. I expected one of the geniuses in the crowd to quickly shout out "16,777,216". The actual answer that I got was "412". This type of converter has been called the delta-sigma (generally East Coast, USA) or the sigma-delta (generally West Coast, USA), but the end result is the same regardless of the name assigned to the circuit. It offers higher accuracy and lower quantization noise through the digitization process than ever before. For instance, one of the most recent converters on the market has a 640-mVp-p input range. With a 640-mV input, 24 bits equates to a 38-nV LSB size (with accuracy to better than 21-bits rms). Is this finally true competition for the purely analog domain circuits? Quite frankly, yes. These converters are neck-to-neck with some of the best. And now, in the mixed-signal domain, just as in the pure analog domain, noise levels (effective resolution) versus bandwidth (data rate) are presenting similar design challenges. Noise reduction and the analog designerýs layout challenge If you want to design a good low-noise analog circuit, layout strategies could make or break you. In analog circuits, power supply connections should be bypassed with good high-frequency capacitors positioned close to the device pins. In most cases 0.1 microF ceramic capacitors are adequate. All of the analog circuits should be powered directly from the power supply outputs to avoid a mix-up with the digital transients. Power supplies should be low noise, which requires sound engineering and careful filtering. If the circuit trace interconnections are short, they are less susceptible to conducted and radiated noise. Particular attention should be paid to the operational amplifier's high-impedance summing junctions. Digital circuit traces can be a real nuisance when in the vicinity of analog circuit traces. Coupled glitches are hard to eliminate short of rethinking the layout. Designing in fast logic is not recommended (assuming TTL or CMOS logic). The fast edges are more prone to emit or couple unwanted signals into sensitive areas of the circuit. The return of the separate power planes The discussion above should sound familiar, but let's look at these issues from the mixed-signal, high-resolution (21-bit accurate) perspective. The best layout approach is to power the analog part of the ADC from one supply and the digital from a separate +5-V supply. This is a change. A few years ago, the recommendation was to put the entire converter on the analog plane. As usual, good decoupling practices should be used on both the analog and digital supplies. A 1- to 10-microF capacitor in parallel with a 0.1-microF ceramic capacitor is usually recommended. For either supply, high-frequency noise will generally be rejected by the digital filter except for integer multiples of the modulator frequency. Where did the digital filter come from? In the digitizing system, the anti-aliasing analog filter has saved many designs from a noise-ridden disaster. It serves the purpose of rejecting high-frequency (uninvited) noise in the analog system so that the digitizer doesn't alias unwanted signals into the bandwidth of interest. One would assume that the anti-aliasing filter would always be a permanent fixture, placed before the ADC. With the delta-sigma converter, the internal digital network has nearly replaced this analog function. Delta-sigma ADC manufacturers are proudly promoting the simple R/C low-pass filter at the input of the converter as the answer to all anti-aliasing problems. In fact, this filter does provide a small amount of high-frequency attenuation, but that is not the primary function of this simple low-pass filter. The most disruptive noise signals that are present at the input of the ADC are the switching currents coming in and out of the converter itself. The first stage of the unbuffered delta-sigma converter is fundamentally a switched-capacitor network. Switching glitches can easily be measured with a few hundred ohms on the inputs and an oscilloscope. This glitch energy can disrupt the measurement of the small voltages at the inputs by upsetting the driving input circuitry. A step toward solving this problem is to place this R/C filter on the inputs of the converter. Additionally, if the device has differential inputs, a 0.1-microF capacitor can be placed directly across the inputs. This is done to attenuate high-frequency noise that is present at the input pins of the device. Note that this technique is not recommended for analog operational amplifiers. Stay tuned: the closure of the analog and digital domains The discrete, high-precision, analog front end in the data-acquisition circuit is not outdated, but is being gently pushed further into its exclusive corner. This is not to say that the demand for precision data-acquisition circuits has disappeared. The solutions to these problems are changing in orientation from the analog-dominated circuit to digital. Although the task of signal processing seems to be migrating to the digital domain, good analog engineering practices still apply. To my delight, it seems that the art of analog hardware design may be renamed but will never be obsolete. Analog Main | Product of the Week | Columns | Editorial | Tech Notes
|
|||||||||||||||||||||||||||||||||||
|
Copyright © 2003 ChipCenter-QuestLink About ChipCenter-Questlink |
||||||||||||||||||||||||||||||||||||