AI is powerful — but it performs at its best when you give it context. That one word, context, is the difference between a vague answer and a useful one. I love analogies, so let me start with a story that perfectly illustrates how AI works, why it sometimes gets confused, and how a little backstory can completely change the outcome.

I have a lot of interests. My wife never knows what topic I’m going to jump into next. Add a splash of ADD and maybe a dab of autism, and you’ve got a recipe for delightful chaos. One minute I’m talking about antennas, the next minute I’m redesigning a circuit board, and somewhere in between I’m wondering where I left my digital calipers. Humor aside, this unpredictability is exactly what happens when people interact with AI without giving the backstory. When context is missing, the conversation feels scattered — and the results can be just as confusing.

Here’s the first thing I want folks to understand: AI is not Google. Search engines look for keywords; AI looks for understanding. Most people still type prompts like search queries, then wonder why the answer feels incomplete. The truth is simple — AI needs context to deliver meaningful, accurate, and helpful responses.

Sharing My Experience Via Context

I use AI every single day — for work, for hobbies, and for some very technical challenges. I’ve experimented with multiple AI platforms and models, and even when using highly capable systems, I’ve learned that they still need strong context. Sometimes the AI gets things wrong, but more often than not, it’s just doing the best it can with the context it was given. When I improve the context, the answers improve. It’s that straightforward.

Let me ground this in something real. I’m building a seven-band HF SSB transceiver — a HAM radio designed to communicate across the globe. My background is in electrical engineering with a heavy focus on digital systems and assembly language. Communications theory? We touched it, but building an entire HF transceiver from scratch is a different level of complexity. That’s where AI became my technical co-pilot.

At the beginning of this journey, my questions were broad. I’d ask something like, “How do I design a mixer stage?” and the answers were generic. Once I started adding backstory — my frequency range, my architecture, my available components, my goals for stability and cost — everything changed. The AI moved from giving textbook answers to giving practical guidance that matched my actual project. Same AI, different context, dramatically better results.

As a licensed HAM, I passed exams that covered electronics fundamentals, but not to the depth required to design an entire transceiver. AI has helped me bridge that gap by turning curiosity into structured learning. Sometimes it even answers questions I didn’t know how to ask — because the surrounding context helps it anticipate what I’m really trying to achieve. And when it makes a mistake? I challenge it. I provide more context. I say, “Nope, that doesn’t fit this design,” and suddenly the next answer aligns much closer with reality.

Think about it this way. If I ask my wife, “Honey, have you seen my digital calipers?” she might give me a blank stare because the context is missing. But if I say, “Honey, I’m trying to measure the distance between clips on my PCB so I can build an RFI enclosure — have you seen that silver measuring tool with the digital readout?” she instantly understands the context and points me to the dining room table… along with a reminder that it belongs in the garage. Same question, more background, completely different outcome.

That’s exactly how AI works. Context unlocks clarity. Context shapes relevance. Context reduces guesswork.

My goal here isn’t just to help you find your car keys or troubleshoot a circuit. My goal is to help you understand how to think when you interact with AI. The old saying still applies: “Garbage in, garbage out.” But today I’d rephrase it slightly — “Weak context in, weak results out.” When you give AI richer context, you’re not just asking a question; you’re building a shared understanding that leads to better collaboration.

Some people worry that AI will replace them. I don’t see it that way. AI is a tool — a very powerful hammer — but without context, every problem still looks like a nail. The real advantage belongs to the people who learn how to use context effectively. They’re the ones who get clearer answers, faster solutions, and deeper insights.

So here’s my elevator pitch: If you want better results from AI, don’t just ask questions — give context. Explain what you’re doing, why you’re doing it, what you’ve already tried, and where you’re stuck. Treat AI less like a search bar and more like a collaborator who needs backstory to help you succeed.

Because at the end of the day, you’re not going to be replaced by AI. You’re going to be replaced by someone who knows how to use AI with the right context — and knows how to swing that new hammer with confidence.

Try it. Use it. Tell it it’s wrong. My two favorites, Copilot and ChatGPT.

My About page provides the background of my project, the Freedom7 HF Transceiver.

If this story resonates, comments are welcome. You can also reach me at david [at] kr4bad-dot-communications. no com

And if you believe understanding matters more than black boxes, you can subscribe to my WordPress https://kr4bad.com/?subscribe=1.

73 KR4BAD David

A not-so-linear approach to creating a quality low noise amplifier for my HF transceiver.

I am very aware of AI, what it can do, how it works, and how it can let you down. I use it daily as an assistant as a professional IT architect. I took electrical engineering in college but my emphasis was digital logic and microprocessors. As an amateur radio operator, I have a great interest in how all these radios actually work. Some understanding is gained by study for the FCC license exams. I’m currently a General (second-level) operator. I am however, studying for my Extra (highest-level) license. This ticket allows more areas of the band spectrum that are off-limits to the lower-level licensees. Recently I’ve gained a lot of my detailed HF understanding using Microsoft Copilot’s AI during the design of components for my DIY HF transceiver the Freedom7.

I will take the reader through my design process here now and also share that I’ve already done this with failures. This is the not-so-linear aspect I referred to in my subtitle above. I’m currently working on a 20 meter band-pass filter for the second time. This post will describe my low noise amplifier (LNA) that comes after the band pass filters in the receive chain. The LNA is shown in the diagram below.

I have already designed switching mechanisms for 7 band-pass filters, for 10, 15, 17, 20, 40, 80, and 160 meters that are not shown in this diagram, but it shows the need for amplification after the passive band-pass filter to provide a serviceable signal at the mixer.

I designed this LNA already and I’ve even modeled it using LTSpice. I have a schematic in KiCAD but I left it and moved to LTSpice because I wasn’t satisfied with the design and wanted to model it’s behavior and understand it’s performance. Currently the best design that I have can be shared with the reader via the schematic in LTSpice.

To add to my not-so-linear wording in my subtitle, I’ll describe another failure (lesson-learned?) at this point in my design process here. I have a nice box with a wide range of ceramic capacitors and also electrolytics. I also have a cheap box of radial inductors with a wide selection of values. I modeled this on a breadboard. I think C3a should be an electrolytic (not shown). I broke out the 12v power supply, fired up the signal generator, and connected a 50 ohm connector to my oscilloscope. I expected beautiful results. Oh, how I was wrong!

With the help of Microsoft’s Copilot, I’m now going to explain this failure and the newfound education I’ve acquired.

When My 1–30 MHz LNA Met a Breadboard: A Cautionary Tale

Designing the LNA on paper felt elegant. Simulating it felt even better. Then I built it on a solderless breadboard — and everything fell apart. The amplifier oscillated, detuned itself, picked up every stray signal in the room, and generally behaved like it had a personal grudge against me.

So what happened? The short answer: breadboards and RF don’t mix. The long answer is much more interesting.

A solderless breadboard looks electrically simple, but at RF it’s a jungle of unintended components.

1. Breadboards Have Enormous Parasitics

• Each tie point adds 2–5 pF of stray capacitance.
• Each row has tens of picofarads between adjacent rails.
• Every jumper wire adds tens of nanohenries of inductance.

At audio frequencies, these parasitics are irrelevant. At 1–30 MHz, they’re circuit‑destroying.

Your carefully tuned input network suddenly becomes a random LC filter. Your bias network becomes a resonator. Your transistor sees a completely different impedance than you designed for.

2. The Ground Plane Doesn’t Exist

RF circuits need a solid, low‑impedance ground plane. Breadboards offer the opposite:

• Long, thin ground rails
• High inductance
• No shielding
• No controlled return paths

The result? Your LNA’s ground reference floats, shifts, and radiates. The amplifier starts behaving like a tiny radio transmitter — and a very bad one.

3. Oscillation Becomes Almost Guaranteed

Wideband LNAs are inherently sensitive. They need:

• Short leads
• Tight layout
• Controlled impedance
• Proper decoupling

A breadboard gives you:

• Long leads
• Random layout
• Undefined impedance
• Decoupling capacitors connected through inductive rails

This is the perfect recipe for VHF oscillation, even if your design is only meant for HF. Many LNAs will happily oscillate at 80–200 MHz if you give them the chance — and a breadboard gives them every chance.

4. The Breadboard Acts Like an Antenna Farm

Every jumper wire is an antenna. Every row is a transmission line. Every gap is a slot radiator.

Your LNA ends up amplifying:

• Local AM broadcast stations
• Switching noise
• Your laptop’s USB emissions
• The fluorescent lights
• The neighbor’s lawnmower ignition noise

Instead of a clean 1–30 MHz signal, you get a chaotic RF soup.

5. Power Supply Noise Goes Straight Into the Amplifier

Breadboards provide almost no isolation between:

• Power rails
• Signal rails
• Ground returns

Even a clean bench supply becomes noisy once it hits the breadboard. Your LNA sees that noise as input signal and amplifies it gleefully.

The Takeaway

The LNA design wasn’t the problem — the construction method was. Breadboards are fantastic for digital logic, microcontrollers, and low‑frequency analog. But for RF, especially wideband RF, they’re essentially parasitic component generators.

If you want an LNA to behave, you need:

• A copper‑clad ground plane
• Short, direct connections
• Proper shielding
• SMD components if possible
• A PCB or at least “Manhattan style” construction

My LNA didn’t fail because the design was bad — it failed because I built a precision RF amplifier on a device that’s basically a parasitic capacitor farm. Lesson learned: breadboards and megahertz don’t mix.

Decisions, Decisions: A Story to Explain

At this point in the LNA design process, do I proceed with the PCB manufacture? Do I add Harwin RFI shielding clip pads to the PCB? Am I really ready to pull the trigger on the PCB design using the schematic I’ll update in KiCAD from LTSpice?

My decision was to go through deep discussion with AI today and write this blog post, now describing my design of the LNA schematic in great detail. My post is aptly titled “Revolutionary New Design Process Includes AI” and while I want to re-validate my design, I’m going to provide the guidance I was originally given here in this post. This was more than just guidance, it was education. It really enlightened my understanding of the overall design process.

Designing a Low Noise Amplifier for HF: A Story to Sell Teamwork

I am certainly humbled by the power of AI and as a manager of people, I know the value of teamwork. I’m going to show the reader the value of using AI to answer all of the hard questions. At this point in time, my goal is to design an LNA for my HF radio that just works. My goal is described as follows:

Amplify very weak RF signals from the antenna with:

• Low noise figure (NF)
• Reasonable gain (10–20 dB is typical at HF)
• Good input/output matching (usually 50 Ω)
• Stability (no oscillations, no weird behavior with strong signals)
• Linearity (doesn’t distort when a strong nearby signal appears)

At HF, the first active device largely sets the receiver’s noise performance. Everything I do in this stage—biasing, matching, layout—will feed into that.

My device has already been chosen. I have a supply of BFR93A SMD bipolar junction transistors on-hand. I think of this transistor as a transconductance device: input voltage → output current, then a load resistor/transformer turns that into voltage gain, hence the term amplifier. My LNA amplifies AC or RF signal input.

DC Bias

First and most importantly, I need to set the operating point of the transistor or amplifying device with DC bias. The transistor must sit at a quiescent point (Q‑point) where:

• It’s in the region where noise performance is good
• It’s in Class A (conducts over the full RF cycle)
• It has enough headroom for signal swing

I would typically set:

• Collector current $I_C$: often 1–10 mA for HF LNA
• Collector voltage $V_C$: maybe half the supply (e.g., 6 V on a 12 V supply)
• Base bias via a resistor divider or current source

Basic idea:

• Choose $I_C$ for a good trade‑off between noise, gain, and power
• Use a collector resistor $R_C$ such that:

$$V_C\approx \frac{V_{CC}}{2}=V_{CC}-I_C\cdot R_C$$

• Base bias network sets $V_B\approx V_E+0.6\text{–}0.7\text{V}$

Emitter Bypassing

In a Bipolar Junction Transistor (BJT) amplifier, the emitter resistor sets the transistor’s bias and stabilizes the gain. That resistor also introduces negative feedback, which keeps the amplifier linear and predictable, specifically needed for this sensitive amplifier.

But there’s a catch.

Negative feedback reduces gain — including the gain you want at RF.

Emitter bypassing is the technique of placing a capacitor in parallel with the emitter resistor so that:

• DC still sees the resistor (for stable bias)
• AC/RF sees the capacitor (which looks like a short at RF)

In other words:

The resistor controls the transistor at DC. The capacitor “turns off” the resistor at RF so you get more gain. See the balance here?

If you return to my schematic, I have 3 components here, R4, C3a, and C3b. My 10uF capacitor, C3a is electrolytic as a choice because it’s suited for the low frequency short.

Emitter bypassing is powerful, but it comes with trade‑offs. The emitter resistor provides negative feedback that helps prevent oscillation but by bypassing it, you remove that feedback at RF, reducing stability. This design choice affects all that bad stuff I talked about above. I’ll have the option of changing values on my PCB using SMD packaging when I’m at that point.

I’ll wrap up this section with my best explanation. In a low‑noise amplifier, the emitter resistor sets the transistor’s bias and keeps the circuit stable, but it also reduces gain and adds noise. By placing a capacitor across that resistor, we “bypass” it at RF frequencies. The transistor still sees the resistor at DC, so the bias stays stable, but at RF the capacitor looks like a short, restoring gain and improving the noise figure. The trade‑off is that bypassing removes some of the stabilizing feedback, so layout and grounding become more critical to prevent oscillation.

Input and Output AC Coupling

As RF is the signal we’re processing we need input and output AC coupling at both ends of the amplifier. We will use capacitors to handle this job. The input capacitor will block DC so that our biasing is unaffected. It will also pass our RF with proper reactance, really around 50 ohms at the lowest frequency of the amplifier’s bandwidth. I choose its value so its reactance at the lowest frequency of interest is small compared to 50 Ω:

$$X_C=\frac{1}{2\pi fC}$$

For example, at 1.8 MHz, you might want $X_C\ll 50\mathrm{\Omega }$, so $C$ in the hundreds of pF to a few nF. I think it’s currently 100nF and that’s subject to change. Again, notice the not-so-linear approach to all this.

The output coupling capacitor needs to do the same. Its value was chosen so its reactance is small at the lowest frequency of interest and it blocks DC from the collector/drain. It will pass RF to the next stage, the mixer.

Impedance Matching

Now we’re seriously going into uncharted territory. It’s also where this RF stuff seems like an art in itself. I want to start with the input matching. I want the source (antenna and passive filter, usually 50 Ω) to see a good match into the LNA input.

Input Match

But the transistor’s input impedance is not 50 Ω by default. It’s some complex value depending on:

• Bias point
• Device parameters (gm, β, capacitances)
• Frequency

Realize at this point, we’ll interject a change between the true input and the transistor device. I would use a matching network:

• L‑network (series L, shunt C or vice versa)
• Transformer (turns ratio to transform 50 Ω to the desired impedance)
• Tapped inductor or autotransformer
• Band‑pass network (for selectivity + matching)

I’ll use an LC network because I already have a coupling capacitor that can partner in a series LC situation to give us the impedance close to 50 ohms.

Copilot AI helped me realize there’s no perfect solution here and this is yet another piece of the LNA puzzle that may need some trial and error component value changes. The impedance that gives minimum noise figure is often not the same as the impedance that gives maximum gain for the amplifier.

So one would choose a source impedance $Z_S$ (via matching network) that:

• Is close to the optimum noise impedance of the transistor at that frequency
• Still gives enough gain and reasonable match

This is one of the core design tensions in an LNA. And, I need to get the datasheet and review my current schematic before I can give it final approval.

Output Match

In keeping with the broadband 1-30MHz concept, we can’t use any form of LC combo. And, as I’m writing this .. . I gave AI this prompt: “Now, I’m questioning a 100nF, 47uH LC series input matching combo.” It seems that my design for the AC coupling was satisfactory but the LC combo to get the impedance matching is off.

Not-so-linear is becoming my mantra. I have conveyed my methods and how I use AI to answer my questions and then I tailor more questions as AI educates me. And, I have found errors in my design at this point and I have not finished the full outline of the LNA design aspects.

Conclusions

I’m going to conclude the details here but I am going to share an outline of how this should go when the final design is solidified. If I write about “every aspect” of my HF LNA, a nice structure might be:

• Measured vs. simulated performance (gain, NF, IP3, stability)
• System role of the LNA in an HF receiver
• Choosing the transistor and topology
• DC bias design (with example calculations)
• AC small‑signal model and gain derivation
• Input and output matching (with Smith chart examples if you like)
• Noise figure and how design choices affect it
• Stability analysis and practical stabilization tricks
• Power supply decoupling and RF chokes
• Layout and construction details (lead length, shielding, grounding)

I tend to think that I will use this outline and write about each when my not-so-linear approach at this LNA is complete.

---

My About page provides the background of my project, the Freedom7 HF Transceiver.

If this story resonates, comments are welcome. You can also reach me at david [at] kr4bad-dot-communications. no com

And if you believe understanding matters more than black boxes, you can subscribe to my WordPress https://kr4bad.com/?subscribe=1.

73 KR4BAD David

For years, surface‑mount devices (SMD) carried a reputation for being “too small,” “too fiddly,” or “only for factories.” Many Makers, HAM radio tinkerers, and DIY electronics hobbyists stuck with through‑hole parts because they felt safer and more familiar.

But the truth is this: with a little practice, a bit of flux, some liquid solder, and an inexpensive rework station, SMD work becomes not only approachable—it becomes fun. Once you learn the technique, you’ll wonder why you avoided it for so long.

This article is an invitation to give SMD a chance. You don’t need a professional lab. You don’t need a microscope. You don’t need a thousand‑dollar hot‑air station. You just need curiosity, a steady hand, and the willingness to try something new. I also use these glasses too. $20 on Amazon Yoctosun Magnifying Glasses with LED and headband

Why SMD Is Worth Learning

SMD components offer real advantages for home projects, especially as electronics continue to shrink and more parts become surface‑mount only.

1. SMD Saves Space—A Lot of It

Through‑hole components take up board area on both sides and require long traces to reach their pads. SMD parts sit flat on the board, allowing:

• Smaller PCBs
• Cleaner layouts
• Shorter signal paths
• More room for connectors, inductors, relays, and other bulky parts

In HF radio projects, this matters. You can reserve precious board space for the components that must be through‑hole—like toroids, high‑power RF transistors, or large electrolytic capacitors—while using SMD for everything else.

2. Many Modern Components Are SMD‑Only

If you want access to the latest ICs, filters, op‑amps, microcontrollers, and RF modules, SMD is often the only option. Learning SMD opens the door to:

• Better performance parts
• Lower noise amplifiers
• Modern RF front‑end chips
• Compact voltage regulators
• High‑quality ceramic capacitors

You’re no longer limited to whatever through‑hole parts are left in the catalog.

3. SMD Can Actually Be Easier to Solder

This surprises people, but it’s true.

With through‑hole parts, you often fight with:

• Leads that don’t fit
• Pads that lift
• Components that fall out while flipping the board
• Excessive heat needed for large pins or ground planes

SMD, on the other hand, rewards technique over brute force. With flux and liquid solder, the surface tension does most of the work for you. Pads pull the solder into place. Components self‑align. Mistakes are easy to fix with hot air.

4. Rework Is Faster and Cleaner

A cheap hot‑air rework station can remove surface mounted ICs in seconds. Try doing that with a 40‑pin DIP without damaging the board.

For prototyping and experimenting, SMD is incredibly forgiving.

What You Actually Need to Get Started

You don’t need a professional setup. A beginner‑friendly bench can be built for less than the cost of a single high‑end soldering iron.

Here’s a realistic starter kit:

• A basic hot‑air rework station (the inexpensive ones work fine)
• A small‑tip soldering iron
• Liquid solder (solder paste or low‑melt solder works great)
• Good flux (this is the real secret weapon)
• Tweezers
• A magnifier or cheap USB microscope (optional but helpful)
• Remember my glasses above

That’s it. No ovens, no fancy stencils, no industrial equipment.

Where Surface-Mount Fits in HF Radio Projects

HF radio designs often mix SMD and through‑hole parts. Not everything belongs has to be surface mounted, and that’s perfectly fine.

Great Candidates for SMD in HF Projects

• Bypass and decoupling capacitors
• Op‑amps and low‑noise amplifiers
• Filters and matching networks
• Microcontrollers and logic ICs
• Voltage regulators
• Small RF transistors
• Resistors and small inductors

These parts benefit from short leads, low parasitics, and compact placement.

Better Left as Through‑Hole

• Toroids and large inductors
• High‑power RF finals
• Large electrolytic capacitors
• Connectors and mechanical components
• Heat‑dissipating devices that need bolted heatsinks

Using SMD where it makes sense frees up board space for the components that must be larger or mechanically robust.

The Learning Curve Is Real—But Short

Your first few attempts may feel awkward. Components may fly off your tweezers. You might bridge a few pads. You might overheat a resistor or two.

But then something clicks.

You learn how much flux is “just right.” You learn how solder paste behaves under heat. You learn how to nudge a part into place and let surface tension finish the job.

And suddenly, SMD stops being scary and starts being empowering.

Why You Should Try SMD on Your Next Project

If you’re a Maker, a HAM, or a DIY electronics enthusiast, SMD opens up a world of possibilities:

• Smaller, cleaner, more professional‑looking boards
• Access to modern components
• Faster assembly and rework
• Better RF performance
• More efficient use of PCB space

Most importantly, it expands what you can build at home.

SMD isn’t just for factories anymore. It’s for anyone with a soldering iron, a bit of patience, and the desire to push their skills forward.

Give it a try. You might discover that the “tiny parts” are not the enemy—they’re the gateway to better, more capable homebrew electronics.

---

My About page provides the background of my project, the Freedom7 HF Transceiver.

If this story resonates, comments are welcome. You can also reach me at david [at] kr4bad-dot-communications. no com

And if you believe understanding matters more than black boxes, you can subscribe to my WordPress https://kr4bad.com/?subscribe=1.

73 KR4BAD David