3 dB bandwidth

The Decibel is a Unit of Attenuation, defined in terms of the ratio of power levels only.

Attenuation = 10 * log10 (Pin/Pout) in Decibels.

If (and only if) the input/output impedances are identical, then the Attenuation (in Decibels) can be calculated as the log10 ratio of the *square* of the currents or voltages.

Or, Attenuation = 20 * log10 (Vin/Vout) in Decibels. /|\\ |____ Notice where the squares went to?

That calculation using (Vin/Vout) produces units of dB only when the input/output impedances are identical.

For example. An audio power amplifier requires an input of 1mW into 600 ohms, for an output of 1W into 15 ohms. The voltage/current ratios are 0.77/3.87 and 1.29/258.

Only the power ratio produces legitimate units of dB.

My point is, if you don't know fundamentals, what good is it showing circuits?

To wit, look at all the folderol that continues here on these NG's about blinking LED's. Not once have I observed a calculation, just things like, "I took off two turns (on the inductor) and it started working". No one has a clue WHY. That's NOT design... that's hacking.

...Jim Thompson

I'm not sure but it isn't electronics as you and I know it !

Some yrs ago I was the liaison guy between my client ( a UK audio manufacturer ) and Salford University who run a specialist electronic course in the electro-acoustic area. We ended up offering them a one-year placement for one of their students since they like to keep industry ties active. Indeed we were going to repeat the exercise to but then the company was placed in a situation where it had to cease trading.

Whilst I was at Salford, I saw some of their labs and work assignments. One was a mic pre-amplifier. I had to point out to the lecturer who was their liaison guy with me that it was actually an instrumentation amplifer they were learning about !

I had other occasions to criticise the student's level of relevant tuition ( but the student himself was pretty ok I should add ). This lecturer sighed and admitted their electronics lecturer was more into heavy electrical so what did you expect ?

I doubt that the skill of discrete design is taught at all these days.

Graham

Thanks, Tony.

As you point out, the calculation using voltage is identical to the calculation using power:

We can examine the -3dB point in a RC low pass filter. This gives the half-power point as well as a phase angle of 45 degrees, so it is a useful and meaningful parameter.

Although the source impedance changes as the frequency is varied, we assume the scope or voltmeter has negligible loading on the signal.

The voltage ratio is then 1/sqrt(2), which gives a power ratio of

(1/sqrt(2))^2 = 0.5

so

and

As you point out

Since power is voltage squared, the two equations are identical and give identical results.

So the legitimate unit for voltage calculations is the dB.

Mike Monett

None whatsoever, except for those who argue to hear their head resonate ;-)

...Jim Thompson

It's just *attenuation* now is it ? Snip rest of trolling garbage.

Holy S*it !

Graham

Graham,

Please be a little more tolerant. As you know, gain can also be understood as a form of attenuation. In other words, gain = - attenuation

Tony Williams is one of the oldest members of sed, and he rarely gets involved in stupid arguments such as this. If you look through the archives, you will find his contributions are often quite brilliant and certainly far above the general level we now experience in this newsgroup.

Tony is not a troll, and it is ungenerous to refer to his post in such a manner.

Mike Monett

Have a look at "Radio Designer's Handbook" by F. Langford-Smith (first published 1934). In my 4th Edition copy (1953) the whole of Chapter 9 is devoted to the subject of decibels.

It is quite clear that only the power ratio is directly described by decibels, if you wish to express the ratio of other electrical quantities in decibels you "must involve the resistance". (Page 807)

I have no idea, I did my training in the 1960s.

So you have a differential signal? Does it have some bias return point, or totally floating and you have to establish common-mode?

...Jim Thompson

I guess I'd have to turn the argument around and say the dB is dimensionless. If you want the result in power, use 10*log. If you want the result in volts or amps, use 20*log.

Assuming the same impedance, the results are identical. They refer to the same thing. Here's a small table:

dB Volts Watts -3dB 1/sqrt(2) 0.5 -6dB 1/2 1/4 -10dB 1/sqrt(10) 1/10 -20dB 1/10 1/100 -40dB 1/100 1/10000 -60dB 1/1000 1/1000000

You can also tie the dB to a specific unit. For example, dBm refers to milliwatts, and dBV to Volts. In these examples, the dimension is defined by the unit.

So where's the problem?

Mike Monett

Thanks, Jim. It's amazing how these threads can go on forever and not solve anything. But when you entered the fray, hopefully that will put a stop to the nonsense. Only an idiot would be foolhardy enough to argue with you.

Let's get back to design problems. I have a problem. I want to make a very high impedance buffer to drive a so-called 24-bit adc (actually only

The problem is I have not been able to find a suitable op amp with bias currents in the fA region, and common mode rejection good enough to use with a 19 bit adc.

For example, the LMC6482 is spec'd at around 20fA, but has only 80dB or so common mode rejection. I believe this is only good enough for 12 bits.

Do you have any suggestions on the perfect op amp to use?

Mike Monett

Basic skills are fine, and needed of course, but good as they are, they don't carry one far. Refining those basic skills to teach how engineers think about designing, organizing and getting a circuit to do the job at hand, analyze and evaluate it, choose a better one, or a better component, etc., these are badly needed skills that most engineers don't get exposed to in college, and perhaps not at work later either. So it's worth teaching. Actually, Jim, MIT used our book for many years in its EE graduate-level instrumentation course. That's not the audience we had in mind for our book, but it does illustrate the need one teacher at MIT saw for exposure to our type of material at some point in one's engineering education.

A common technique in the old days was to create a set of floating supplies for the input amplifier. Run your LMC6082 or LMC6062 op amps (these are better choices) on 3V regulators, etc., whose common reference terminal operates from the opamp's output. This way there won't be any common-mode input error.

Yep, PhD's need that sort of instruction... that's why I was hired as a tech by Woodson/Jackson/Melcher... the PhD candidates were running around trying to screw a fuse up their ass ;-)

...Jim Thompson

--------------------------------------------------------

Adrian Tuddenham wrote:

Don't you consider it somewhat foolish to consider one to the exclusion of the other though ? *Both* are required IME to do a decent job. No point re-inventing the wheel constantly. Indeed many or even most practical circuits have been around for decades.

The really nice bit is tweaking a popular configuration to enhance performance.

Graham

I'd like to duplicate the performance of the HP3456 6-digit dvm as much as possible. It has floating inputs and can do 4-wire ohms measurements. I have a bunch of AD7791's but can easily change if needed. They have 20ppm linearity, which is good enough to start.

The problem is input impedance. It needs a buffer to get the 10 gazillion ohm input impedance on the low dc ranges. I expect to use an instrumentation amp configuration, but the problem is common mode linearity. From the LMC6482 datasheet:

LMC6482 CMOS Dual Rail-To-Rail Input and Output Operational Amplifier September 2003

Low power, single supply data acquisition system solutions are provided by buffering the ADC12038 with the LMC6482 (Figure 14). Capable of using the full supply range, the LMC6482 does not require input signals to be scaled down to meet limited common mode voltage ranges. The LMC4282 CMRR of 82dB maintains integral linearity of a 12-bit data acquisition system to +/-0.325 LSB. Other rail-to-rail input amplifiers with only 50dB of CMRR will degrade the accuracy of the data acquisition system to only 8 bits.

I don't know how they translate 82dB of common mode rejection into 12 bits of linearity, and their story is complicated by the input shown in Fig. 14. This is an AC-coupled input with Cin = 1nF and Rin = 500k. So their story is somewhat muddled.

But it seems reasonable that cmmr will translate into integral nonlinearity, and I'd like to reduce that as much as possible to avoid degrading the performance of the AD7791. A google search on the translation mechanism yields nothing as yet, so I don't have any equations to offer.

Any hints on solving this problem?

Mike Monett

Thanks Fred. That device was not in my database. Unfortunately, it looks like the maximum input impedance it can work with is 100k before it starts developing significant errors. That is supported by the input leakage spec of +/- 10nA.

So it's not really suited for the application I have in mind. But thanks for your consideration.

Mike Monett