Voltage Variable and Digital Step Attenuators

A voltage variable attenuator takes one analog control voltage and delivers a continuous attenuation between its minimum and its maximum. A digital step attenuator takes a binary word and delivers one setting out of a fixed number of discrete ones. The first has unlimited resolution on paper and inherits everything riding on its control line. The second holds a setting that does not move unless the code moves, with a resolution fixed by the sections inside it. Neither is the more accurate in general. They fail differently, and a chain is built around the failure it can live with.

A control loop does not care which philosophy a datasheet prefers. It cares whether asking twice for the same loss returns the same loss, and whether asking for more returns more.

What a control voltage actually controls

Attenuation against control voltage is not linear in volts and not linear in decibels unless the designer shaped it to be. Shaping costs circuitry inside the module, so a part that ships with a shaped curve and a part that ships with a raw one are not interchangeable, and the datasheet has to say which it is. A loop that assumes a straight line between two calibrated points is wrong everywhere else on an unshaped curve.

Continuity is also the exposure. Whatever the control line carries moves the loss with it: supply ripple, pickup along the run, drift in the reference that set the voltage. Resolution is therefore not a property of the attenuator alone. It belongs to the attenuator and to the quietness of the board feeding it.

How does a binary word become a step?

The part is assembled from weighted sections that the control word switches in and out in combination. Six sections weighted 0.5, 1, 2, 4, 8 and 16 dB give a six-bit word sixty-four distinct settings. The value printed beside each code is the sum of the sections that code selects. Resolution is then a number you read off the interface: the smallest section, with nothing between two adjacent codes.

Step error comes from the same structure. Each section carries its own residual loss at zero, its own flatness across band, and its own deviation from the nominal value the weighting assumed. Combined codes combine those deviations. How many sections a code switches has nothing to do with how high the code sits: at a carry the higher of two adjacent codes can switch one section where the lower switched five, so two adjacent codes need not differ by exactly the printed step.

Can a control loop close around either of them?

Monotonicity decides it. A step attenuator is monotonic when attenuation never decreases as the code increases. A part can sit inside its accuracy window at every code and still be non-monotonic at one boundary, and a loop asking for more attenuation and receiving less will hunt at that boundary and nowhere else. Accuracy and monotonicity name different defects, and they are routinely confused because both are quoted in decibels.

A voltage variable part has no codes, so the question changes shape. The loop has to know where it sits on the curve, and if the curve is shaped, it depends on the shaping staying put. Repeatability here means returning to the same control voltage and getting the same loss, read across the band rather than taken from one frequency.

Where does each one fail?

A voltage variable part fails in the control domain. A control line noisier than the design assumed, or a reference that has moved since calibration, shows up as an attenuation that is wrong in a way no single figure predicted. Nothing breaks; the loss is simply no longer the loss the system computed.

A step attenuator fails in the section domain. A section that drifts off its nominal weighting produces one code whose loss is wrong while the others stay inside the window. The failure is addressable, which helps diagnosis: the suspect setting can be named, revisited and measured on its own. It is also invisible if the chain only ever uses the codes that happen to be right.

Reading one against the other on paper

What the parts themselves are for is set out in the attenuator primer. Compared here are the five columns a range figure does not settle: how the part is commanded, what its resolution is by construction, what its characteristic failure looks like, whether a setting comes back the same way twice, and what has to be measured before it is trusted.

Two control philosophies against the columns that decide between them

Dimension	Voltage variable	Digital step
Command	One analog voltage, continuous	A binary word, one setting per code
Resolution	Set by the control line and its quietness	Set by the smallest section, fixed by construction
Characteristic failure	The loss stops matching the value computed from the control voltage	One code drifts off while the rest hold
Repeatability of a setting	The same voltage has to be re-established exactly	The same code is reapplied by the interface
What to measure	The full transfer curve across band, not two points	Every code, not a sample of them

What to measure before the figure is quoted

Both parts are two-ports and the bench work is the same in kind. Insertion loss is what the through path costs at the minimum setting and it cannot be zero. Flatness is the variation across the band at a fixed setting, a separate figure per setting. Accuracy is the gap between the loss asked for and the loss delivered. Phase shift with attenuation goes on the same list, because attenuation almost always arrives with a change of phase, and in a multi-branch system those changes add between branches.

The three terms most often conflated stay separate: flatness, accuracy and monotonicity name three different defects. The methods behind both parts are in insertion loss and isolation.

The concrete step is short. Write down the codes or the control voltages the chain will actually issue, measure those and only those, in the order the chain issues them, and record the increment between neighbors. What that short table shows is the answer the range column was hiding.