Using stakeholder, system, hardware and software requirements to build a professional wireless microphone.
In the post below—the last of three transcribed from his Writing Good Requirements workshop, with notes and slides from his presentation deck included—Jama Consultant Adrian Rolufs explains common problems teams go through, and how to avoid them. (See Part II here.)
==Start Part III==
Let’s look at an example product using my audio background. I’m going to take a circuit that goes into a professional wireless microphone—the kind of high-performance microphone you’d see someone on a stage, like a MC or a musician, use.
It’s got to be able to handle a wide, dynamic range, meaning it has to be able to record very loud signals and very quiet signals, all with very high quality, and it’s got to be powered off of a battery so that it can he handheld, meaning the connection to the system will be wireless.
So we’re going to talk about some of the requirements that go into the chip; one of the main chips that goes into a solution like this.
First we’ll start at the market or stakeholder requirements level. Often, they’re called stakeholder requirements because stakeholders can be more than just customers.
In most product development organizations customers have requirements, but internal teams also have requirements.
So if I’m building a chip, for example, I have quality requirements that my quality department is going to dictate, but will also be influenced by the customer’s requirements.
And I probably have a production test organization that has to test every one of these devices as they go out the assembly line.
These devices are going to have requirements concerning what kind of access they need to internal circuits, and what kinds of circuits they need to enable them to test in a timely manner—things like that.
The development team might also have requirements; for example, they need to be able to reuse certain amounts of existing circuitry to stay on schedule, or requirements around data costs.
The point is that what we call stakeholder requirements is really a broad category. It could be anybody who has an influence on the product development.
Let’s look at some examples—these would most likely come from customers—which would be focused on the functionality and performance of the device. I’ve got three examples here: One is good and recommended; two are not.
We’ll start with the first one. Say we need a product that can input a microphone signal, convert a signal with two digital audio signals, have two different gain stages and consume less than 20mA while operating.
This is the sort of thing you’d likely hear directly from the customer but not necessarily the sort of thing I’d want to write down as requirements.
This brings up a couple of issues.
First, I’ve got a bunch of requirements mixed together, so it would be easy to miss something, and also it presupposes a certain solution.
It could be this is the right solution, but it assumes certain solutions, so it’s talking about internal details that are over‑constraining the design team.
The team can come up with a different gain structure that works and achieves the results, but doesn’t use 20dB and 0dB of gain. What’s wrong with that? Why do I need to over‑constrain them?
So those are some of the problems with the first one.
Second, customer X needs a 140dB microphone amplifier with a digital output for less than 50 cents. The microphone amplifier shall be low power.
This is the sort of thing marketing might write, because it’s focused on the customer’s request: They need it at a certain cost and everything should always be low power.
It’s very difficult to actually meet these kinds of requirements.
140dB—well, what is that? That’s just a ratio number; I don’t know what that actually is a measurement of. I need some more specificity around that.
As for 50 cents, you have no idea what the solution is yet so 50 cents may or may not be achievable, but it’s good to know.
And then the last one, low power; that can mean almost anything. Low power in one industry could be high power in another, so specificity around what low power means would be beneficial.
So in that case, the first example is more specific and has more detail—although both of the first two are not very atomic so it would be easy to miss things.
The last example talks about two things.
The first one has a problem statement. I love problem statements because they really tie back to the value the solution can offer. It’s giving me some context around what’s in the market today and what the problem is.
It’s saying in the market there are high dynamic range microphones which transmit digitally, and it requires circuitry that’s expensive and large or high power to obtain the necessary performance.
And from that I know that a solution is out there, but that solution is difficult, hard to use or hard to implement, and it can be expensive and it may or may not provide the necessary performance.
You can see how this helps outline the idea of what kinds of problems I need to solve and where the most value is in design.
So based on this, I would know that hitting the audio performance is important, and getting a small solution size that’s low power is also important; those are the key constraints.
To make that have a specific power consumption, say the solution shall consume less than 75mW while in operation.
Now, the other benefit here is 75mW; it’s an actual power number, whereas in the first example I had a current but without knowing the voltage I don’t what the power consumption is, so that’s also not a great example.
So in this case, the last one is the one I would recommend; it has more constraints and a good set of stakeholder requirements. With that, the design team has a good idea of what their goals are, but they’re not over‑constrained.
Now for the next level of detail: Once we have a set of stakeholder requirements, or at least a draft, we can start looking at system requirements. The system requirements are what we’re actually going to build a product against.
We’re not going to build a product directly against the stakeholder requirements because we could have multiple stakeholders and we need to consolidate their requirements into one set.
Or, certain stakeholders may ask for things that we actually end up not satisfying, but we still know that we can build a successful product.
So that translation from stakeholder requirements to system requirements provides the clarity and explicit decisions around what we’re going to do, what we’re not going to do, and what the actual requirements are for this project.
Now, one of my favorite examples of system requirements is the first one—absolutely nothing—and I see this time and time again.
I can’t count the number of times where I see people skipping the system requirements when they’re building a system.
If you’re an engineer responsible for low-level details, how do you know if those low-level details are the right details? Well, you need system requirements first, so we definitely don’t want to skip this level.
Now, the next one: The solution shall have two differential inputs using instrumentation amplifiers. The instrumentation amplifiers shall be followed by sigma‑delta ADCs. These are really low-level component requirements.
We’ve already jumped to the conclusion that we’re going to have a specific architecture in the hardware. What if part of the solution needs to be software? We haven’t said anything about that and we could already be over‑constraining the design team when some other architecture would be more appropriate.
It could be an instrumentation amplifier is not the best choice. We don’t need to constrain that at this level. So the last example here is really a better example of system requirements.
What about power consumption? It’s going to 20mA while in operation. I said before, current on its own is not necessarily the best example. With this you would typically provide a supply voltage range so then it would become clear.
What about the signal levels? Stating what the signal range needs to be provides a lot of detail around what the architecture of the design needs to be, without over‑constraining it.
And then, the overall end‑to‑end, signal‑to‑noise ratio: 140dB A‑weighted gives me a very clear statement of the overall performance, again without over‑constraining.
So for system requirements I like that last one.
These system requirements are all focused on the performance of the signal path. There should also be some system requirements here that talk about constraints on size, constraints on packaging and things like that.
Now we know we need to build something that consumes relatively low power, takes in a very wide dynamic range signal and maintains the quality of that. So we can start talking about architectures in response to these system requirements.
Let’s say we use a hardware device that has analog to digital conversion with two signal paths, both of which have medium performance but which we can combine together to obtain high performance, and that’s actually the common solution in the application.
And then we use a software algorithm to combine those signals, so we’re going to need a DSP to run the software algorithm and process the signal to output this resulting signal of 140dB A‑weighted signal noise ratio.
Based on that, we can now talk about the hardware‑specific requirements.
Here are some examples of different possibilities. The first one is a block diagram of the architecture.
I’m visual; I love block diagrams. I love schematics because they’re very intuitive. I can relate to them very well. They don’t make good requirements, unfortunately. It’s very difficult to test a diagram. It’s very difficult to make sure you didn’t miss anything in a diagram. So having a diagram on its own is not sufficient.
A really good solution is a diagram complemented by a set of requirements that attach to every important detail of that diagram.
That way, visual people have something to see, but we also have atomic requirements that we can test against and trace to make sure we didn’t miss anything, and also so we can manage changes.
If I make changes to this diagram based on changes to the architecture or customer requirements, it might be hard to actually know what those changes were, whereas if I have individual requirements I can track, I can easily know.
The second example is just a description of functionality, a response to the requirements. This is saying what the signal path of the device is; the architecture is describing a specific part number. This is down in the design descriptions where this belongs. It’s not hardware requirements.
So the last example is one I like for hardware requirements. We’re talking power consumption; we’re getting more specific.
We know that I’m building a chip, the power consumption is going to vary and I want to know what it’s typically at and what its maximum can be, so we’re specifying that.
Again, we’re repeating the input signal level because that input signal level was a requirement on the system that’s also a requirement on the hardware.
There is some duplication, but it’s there to explicitly say that this is a requirement on the hardware. I won’t see a requirement for 17uV RMS to 1V RMS on the software, because the software is never going to know about volts; it’s going to know about digital signals.
So even though there is duplication it’s done to make the decisions and the traceability explicit. So then I have requirements on the specific architecture.
Now that we’re down at the low-level and component requirements, the hardware requirements, we can start talking about specific solutions. We’ve got to get into the details of what the solution is actually going to be.
So in this case, in the hardware requirements, you’ll likely see requirements that dictate a certain solution, but that’s okay because it’s quite likely that the design team is the one writing these requirements, so they’re the right ones to make that decision.
As you probably have guessed by now, the last one is my recommendation for well‑written hardware requirements.
The last example is software requirements.
I see a lot of teams that just skip software requirements entirely and go straight to writing code. It’s really fun to write code, really satisfying, but if you don’t have any requirements, you’re starting without clear directions. We need some requirements.
The second example, some descriptions of functionality, is written as a shall statement. It sounds like a requirement but I’ve got a bunch of stuff mixed together.
I’m talking about two signals. I’m talking about what their performance is. I’m talking about the output. There is too much stuff mixed together here, so the third one is the recommendation: talking about specifics.
I am going to develop this software for a specific DSP, the Tenscilica HiFi 3. It’s going to perform a specific function. It’s going to take two audio signals and it’s going to combine those into one.
(Note: I’d probably need more detail around this. This is probably not enough by itself but I didn’t want to fill the screen with the requirements.)
And then, what’s the sample rate going to be? This algorithm is going to be designed for a specific sample rate or multiple sample rates. It’s an important characteristic of the algorithm. Let’s make sure that’s captured in requirements.
So that is exactly what I would recommend, and in each of these there are a lot more requirements that go along with this. These are just a couple of examples in each category.
Many teams mix up requirements and specifications. It’s very common.
You need to make sure you have a clear understanding of each of them and when to use them. It’s not always easy to decide which one is which, so it’s absolutely critical to have that discussion with your team.
What I see a lot of teams doing is skipping levels of hierarchy, jumping straight from high-level customer requirements down to detailed requirements or detailed specifications. Do that and you’ll have a very difficult time proving that you built the right thing.
It could be you’re operating fast and loose and you’re okay with that. Maybe that’s okay for a very small team in a very small organization. But in every other situation, it’s unlikely that you’d be building something so uncomplicated that you could get away with it. It’s high-risk.
So make sure that you have at least stakeholder requirements, client requirements and some kind of detailed specifications.
That’s the bare, bare minimum for any kind of product. More likely you need more.
What I recommend: Make sure you have a clearly defined process with clear levels of your requirements. If you don’t think you have that, discuss it with your team. What levels do you need? Which one of those diagrams [more in posts Part I and Part II] would be appropriate for your project?
And then there’s the scary question: Do you even use requirements?
Some teams plow ahead without requirements. Think about what kind of problems that can cause:
- Maybe you’ve built products that have not been successful, that maybe needed a late change or maybe even failed, and you had to develop a new product in order to be successful.
- Perhaps you’ve been in a situation where you later learned you missed some important details along the way and realized that you barely got away with it. That’s high-risk too.
When you start with an understanding of the roles different levels of requirements perform, you’re less likely to invite risk and add complications during development, and are much more likely to build the right product.
==End Part III==