How to get all the codes from your high performance SAR a/d converter

The snr of an a/d converter is the ratio of the largest signal it can process to its noise floor. To achieve an snr of up to 102dB, the LTC2379 family specifies a 10Vpp differential input range, which means each inputs swings from 0 to 5V. Somewhere in front of the converter will be an amplifier, whose job is to serve as a good voltage source to charge the converter's sampling caps. The converter's inputs are the amplifier outputs, so the amplifier output must also swing from 0V to 5V. It's easy if you have wide supply rails available; part of your front end may run from ±15V supplies and any op amp running from those rails can swing its output to 0 and 5V. If you don't like to use ±15V supplies and still want to swing from 0 to 5V, you could generate special supply rails just for the last amplifier of, for example -2V and +7V. The reference design for the LT6350 driving the LTC2379-18 does that (see fig 1). The +7V supply is also handy to power your 5V reference. But what if you want to power the amplifiers from one 5V supply rail? You might think that, with a rail to rail op amp, you have just enough room to swing from 0 to 5V, but you don't. Rail to rail output stages are not really rail to rail; they typically get at best to about 10mV from each rail – often with hard clipping and slow saturation recovery times. If good linearity is required, the output voltage should usually stay at least several 100mV from each rail. For example, the LTC6362 (see fig 2) is a low power differential op amp that operates from one 5V supply. The outputs can swing to about 100mV from either supply rail and maintain better than 110dB linearity to within 250mV. If you design your system so your largest signal of interest will not exceed those conditions, you will exercise at least 90% of the converter's codes – within 1dB of the stated dynamic range. In many cases, that's the best solution. It can be comforting to know the amplifier is guaranteed to not overrange (or damage) the converter inputs – it's natural protection. The LTC2379 family features Digital Gain Compression (DGC). With that turned on, the converter will interpret as full scale a voltage swing from 10% to 90% of the reference. So, with a 5V reference, the amplifier output needs only to swing from 0.5V to 4.5V for all 256k codes of the 18bit a/d to be used. Although you get all the codes, dynamic range is reduced a little because the analogue voltage swing is reduced to 8Vpp, although thermal noise remains the same. For the 18bit a/d, quantisation noise is so small that only thermal noise matters; as a result, you lose about 2dB of snr in DGC mode. For 16bit parts, you only lose 1dB of snr in DGC mode because quantisation noise scales accordingly. The single ended (or pseudodifferential) LTC2369 family does not support DGC; an intentional decision because, for single ended unipolar signals, performance near zero is often of most importance. It's when your signal is small that you value fine resolution and low noise performance. For a differential converter, a 'zero' is obtained when both inputs are equal. For a unipolar single ended device, a zero is obtained when the input signal is at ground. For that interface, you need the amplifier to swing to ground. If no external negative supply is available, the LTC6360 – a dc accurate high speed op amp – includes an on chip charge pump that generates a small negative bias voltage, which powers the output stage. This way, the output can swing completely to 0 without coming close to distorting or clipping. On the high side, the LTC6360 can swing to about 4.5V; you could either define that as your largest signal and come within 1dB of full scale on a 5V reference, or use a 4.096V reference and swing full scale. The latter system runs from a single 5V supply, including the reference (see fig 3). Next, turn your attention to input swing limitations. Sometimes, all you want the final op amp to do is buffer the signal into the converter without providing any gain or level shifting. For an op amp in a unity gain configuration, the input swings just as much as the output. Again, if you have wide supply rails available, such as ±15V or -2V/+7V, there is no problem. But if you want to run the op amp from a single 5V supply, it may be tempting to select one of many rail to rail input op amps and believe everything will work. However, rail to rail input stages are composed of two parallel input stages: one that works when the input is close to the positive rail; another that works when the input is close to the negative rail (or ground). Each stage has its own offset voltage and when the signal transitions from one input stage to the other, you get a step in offset voltage at this 'handover' point. This brings a non linearity to the system transfer function. Look at the op amp's datasheet to see if the offset is trimmed in both regimes; if it isn't, the non linearity is probably going to be too bad for 16bit or 18bit integral non linearity performance. The LTC6360, on the other hand, has a tightly trimmed offset across the input operating range and harmonic distortion stays at less than -100dB, even when the signal swings from 0 to 4V, which traverses the handover point – about 3.6V for this part. Another approach to alleviate the op amp's input swing requirements is to use the amplifier in an inverting configuration. LT6350 can be configured with each op amp inverting, so the op amp inputs stay at a dc voltage somewhere in the middle of the supply range. That way, there is no problem with input common mode. A differential op amp, such as the LTC6362, is inherently inverting. When used for single ended to differential conversion as shown, the op amp inputs do swing, but much less than the signal itself. Note that, in each of these inverting configurations, the input impedance to the circuit is resistive, so you need to make sure the preceding circuitry can drive that resistor. Author profile: Kris Lokere is design manager, signal conditioning products, with Linear Technology.