<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0"><channel><title><![CDATA[DB Design RF & Microwave]]></title><description><![CDATA[DB Design RF & Microwave]]></description><link>https://dbdesignmw.hashnode.dev</link><generator>RSS for Node</generator><lastBuildDate>Wed, 18 Nov 2026 13:34:08 GMT</lastBuildDate><atom:link href="https://dbdesignmw.hashnode.dev/rss.xml" rel="self" type="application/rss+xml"/><language><![CDATA[en]]></language><ttl>60</ttl><item><title><![CDATA[Latching vs Failsafe RF Switches: The Decision That Shows Up After a Power Cut]]></title><description><![CDATA[A switch datasheet lists actuation as a line item, somewhere between connector type and operating temperature. It is actually the decision that determines…]]></description><link>https://dbdesignmw.hashnode.dev/latching-vs-failsafe-rf-switches-the-decision-that-shows-up-after-a-power-cut</link><guid isPermaLink="true">https://dbdesignmw.hashnode.dev/latching-vs-failsafe-rf-switches-the-decision-that-shows-up-after-a-power-cut</guid><category><![CDATA[Rf]]></category><category><![CDATA[hardware]]></category><dc:creator><![CDATA[Yang Fei]]></dc:creator><pubDate>Mon, 16 Nov 2026 00:19:48 GMT</pubDate><content:encoded><![CDATA[<p>A switch datasheet lists actuation as a line item, somewhere between connector type and operating temperature. It is actually the decision that determines what your system does at the worst possible moment, and it is usually made by whoever filled in the RFQ fastest.</p>
<h2>Three behaviours, not two</h2>
<p><strong>Latching.</strong> A pulse drives the actuator to the new position and the mechanism holds it there with no current. There are two-coil versions (one coil per position) and single-coil versions that use pulse polarity. Either way, steady-state current is zero.</p>
<p><strong>Failsafe.</strong> A spring or a permanent magnet holds a default position. Applying drive moves the switch away from it; removing drive returns it. Continuous current is required to stay in the non-default position.</p>
<p><strong>Normally open (DCO).</strong> All ports isolated when no drive is applied. Used where the safe state is nothing connected to anything.</p>
<p>The distinction that matters is not mechanical. It is what happens when the 28 V rail disappears at 03:00.</p>
<h2>What latching buys you</h2>
<p>Zero holding current. On a satellite, a portable instrument, or a battery-backed system, that is decisive: a failsafe switch drawing 200 mA continuously is a thermal problem and a power budget problem.</p>
<p>Latching also means the switch does not move on a power interruption, which is what you want when the system is mid-measurement and a supply glitch should not silently reroute the signal.</p>
<p>The cost is state ambiguity. After a power loss and recovery, the controller does not know where the switch is unless it has a position indicator, a stored state, or a homing routine. A stored state is a guess; a position indicator is a specification you should ask for explicitly. Without one, the first action after a cold start has to be a homing cycle, and that costs switching time and consumes life.</p>
<h2>What failsafe buys you</h2>
<p>A defined state under fault. If the safe condition is receiver disconnected or transmitter into the load, a failsafe switch delivers that condition without any software having to be correct.</p>
<p>The cost is continuous current, and a second, subtler cost: the transition itself. When drive is removed, the switch moves. If the transmitter is live at that instant, you have just hot-switched at full power into whatever the default port faces. The default port had better be a load that is rated for it, because that is exactly the moment when the load is asked to absorb everything.</p>
<h2>The failure modes worth designing against</h2>
<p><strong>Coil burn on a latching part.</strong> Driving a latching switch with continuous current instead of a pulse overheats the coil. It is a firmware bug, and it shows up as a switch that works for a month and then stops.</p>
<p><strong>Hot switching on a failsafe part.</strong> As above. Add a hardware interlock that mutes the source before the drive can be removed, or accept that the load is a wear item.</p>
<p><strong>Missing position feedback.</strong> A latching switch with no indicator and no homing routine will eventually disagree with the controller. This is the most common intermittent report I see, and it is never the switch's fault.</p>
<p><strong>Drive polarity on single-coil latching parts.</strong> Reversing the supply to move the switch is fine until a connector is wired backwards during a field replacement.</p>
<h2>How to choose</h2>
<table>
<thead>
<tr>
<th>Requirement</th>
<th>Choice</th>
</tr>
</thead>
<tbody><tr>
<td>Zero standby power, state must survive power loss</td>
<td>Latching, with position indicator</td>
</tr>
<tr>
<td>A defined safe state under fault</td>
<td>Failsafe</td>
</tr>
<tr>
<td>Both: survive fault safe and hold with no current</td>
<td>Latching with a default-position spring (a hybrid, and worth asking for)</td>
</tr>
<tr>
<td>All ports isolated when unpowered</td>
<td>Normally open / DCO</td>
</tr>
</tbody></table>
<p>Two more questions decide the rest:</p>
<ol>
<li><strong>What is the switching time measured from, and does it include the driver?</strong> A latching part quoted at 5 ms is quoting the mechanical transfer. Command-to-settled will be longer.</li>
<li><strong>What does the position indicator report, contact closure or an independent sensor?</strong> A contact closure on the actuator tells you the actuator moved. An independent sensor tells you the RF path moved. They are not the same claim.</li>
</ol>
<h2>Where I sit</h2>
<p>I work on RF switching hardware at <a href="https://www.dbdesignmw.com">DB Design / MeiXun</a> in Wuxi, so this is written from the manufacturing side. We build <a href="https://www.dbdesignmw.com/products/passive-components/coaxial-switches">electromechanical coaxial switches</a> from SPDT to SP12T and DPDT to 67 GHz in latching, failsafe and normally-open configurations, and <a href="https://www.dbdesignmw.com/products/passive-components/waveguide-switches">waveguide switches</a> to 110 GHz where the same three choices apply.</p>
<p>If a switch appears to move on its own, the question to ask first is not about the switch. It is about what the drive line was doing at that moment.</p>
]]></content:encoded></item><item><title><![CDATA[Waveguide vs Coax Above 40 GHz: Where the Crossover Actually Is]]></title><description><![CDATA[Ask five engineers where coax stops being a good idea and you will get five answers: 40 GHz, 50 GHz, 67 GHz, 90 GHz, "it depends". They are all defensible…]]></description><link>https://dbdesignmw.hashnode.dev/waveguide-vs-coax-above-40-ghz-where-the-crossover-actually-is</link><guid isPermaLink="true">https://dbdesignmw.hashnode.dev/waveguide-vs-coax-above-40-ghz-where-the-crossover-actually-is</guid><category><![CDATA[waveguide]]></category><category><![CDATA[Rf]]></category><category><![CDATA[Microwave]]></category><dc:creator><![CDATA[Yang Fei]]></dc:creator><pubDate>Sat, 14 Nov 2026 08:36:10 GMT</pubDate><content:encoded><![CDATA[<p>Ask five engineers where coax stops being a good idea and you will get five answers: 40 GHz, 50 GHz, 67 GHz, 90 GHz, "it depends". They are all defensible, because the crossover is not a property of the cable — it is a property of your loss budget, your mechanical packaging and how often the joint gets mated.</p>
<p>Here is how the decision actually breaks down.</p>
<h2>The physics that sets the floor</h2>
<p>Coax loss rises roughly with the square root of frequency, and then faster than that as the dielectric and the conductor finish stop behaving. A 1.85 mm cable assembly that is 1.5 dB per metre at 40 GHz is close to 3 dB per metre at 67 GHz, and a 1.0 mm assembly is worse again.</p>
<p>Waveguide, by contrast, gets better as frequency rises above its cutoff: WR28 (26.5–40 GHz) has lower loss per metre than any coax at 40 GHz, and WR12 at 90 GHz is dramatically better than 1.0 mm coax.</p>
<p>So on loss alone, waveguide wins above about 40 GHz in every case where the run is longer than a few tens of centimetres.</p>
<h2>The reasons coax still wins at 50 GHz</h2>
<p>Loss is not the only term. Below about 67 GHz, coax often wins on everything else:</p>
<p><strong>Mechanical tolerance.</strong> A waveguide run has to be aligned and flanged; a coax run bends. In a dense chassis, the coax route exists and the waveguide route does not.</p>
<p><strong>Component ecosystem.</strong> Below 50 GHz, coax switches, attenuators, couplers and loads are catalogue items in every form factor. Above 50 GHz the catalogue thins out fast, and above 67 GHz it becomes a custom conversation.</p>
<p><strong>Repeatability of the joint.</strong> A precision coax connector has a defined torque and a defined centre-pin engagement. A waveguide flange has a defined flatness, a defined bolt pattern and a gasket that is easy to omit — and a misaligned flange shows up as a reflection that looks like a component problem.</p>
<p><strong>Test equipment interfaces.</strong> Most VNAs and power sensors present coax. Every waveguide measurement therefore needs a transition, and each transition costs return loss and calibration uncertainty.</p>
<h2>The crossover, stated honestly</h2>
<p>For a switch or a passive component inside a system:</p>
<ul>
<li><strong>Below 40 GHz:</strong> coax, essentially always. The mechanical and ecosystem advantages dominate.</li>
<li><strong>40 to 67 GHz:</strong> it depends on run length and channel count. Short internal runs stay coax (1.85 mm); anything that has to travel, or any path where 1 dB matters, goes waveguide.</li>
<li><strong>Above 67 GHz:</strong> waveguide. The only question is which band and which flange.</li>
</ul>
<p>For a test rack, the crossover moves down: a VNA port at 50 GHz with a 1.0 m run to the DUT is often better served by a waveguide run with two transitions than by a long 1.85 mm assembly, because the assembly's loss and phase drift dominate the uncertainty.</p>
<h2>The cost nobody puts in the spreadsheet</h2>
<p>Waveguide runs bring their own overhead:</p>
<ul>
<li><strong>Pressurisation.</strong> High power in waveguide usually means dry air or SF6, which adds a pressure window, a seal grade and a leak rate to the specification.</li>
<li><strong>Flange inventory.</strong> UG, CPR, CMR, plain and grooved variants are not interchangeable without an adapter, and an adapter in a phase-critical run costs you both return loss and a recalibration.</li>
<li><strong>Band limits.</strong> A waveguide switch is specified per band. A WR28 switch does not cover 26.5 to 40 GHz and 40 to 60 GHz; it covers one of them, and the next band is a different part with a different flange.</li>
</ul>
<h2>A rule that survives contact with a project</h2>
<p>Buy the switch that matches the interface you already have at the DUT. If the DUT is coax to 50 GHz, use a coax switch to 50 GHz and terminate well; do not build a waveguide path with two transitions to save 0.4 dB of loss and lose 3 dB of repeatability in the process.</p>
<p>If the DUT is waveguide — an antenna, a waveguide filter, a high-power amplifier — start from the flange and work backwards. The switch, the loads and the couplers should all be in the same band with mating flanges, and the transitions should exist only where absolutely necessary.</p>
<h2>Where I sit</h2>
<p>I work on RF and microwave switching at <a href="https://www.dbdesignmw.com">DB Design / MeiXun</a> in Wuxi, so I see both sides of this: <a href="https://www.dbdesignmw.com/products/passive-components/coaxial-switches">electromechanical coaxial switches</a> to 67 GHz and <a href="https://www.dbdesignmw.com/products/passive-components/waveguide-switches">waveguide switches</a> to 110 GHz, plus <a href="https://www.dbdesignmw.com/products/active-components/pin-switches">PIN diode switches</a> for the cases where the transition has to happen inside a pulse.</p>
<p>The honest answer to "coax or waveguide" is usually "whichever interface your DUT already has", and the interesting engineering is in making that choice survive the next two years of use.</p>
]]></content:encoded></item><item><title><![CDATA[Reading an RF Switch Datasheet: Five Lines Every Vendor Measures Differently]]></title><description><![CDATA[Two datasheets can both say "isolation: 60 dB" and describe parts that behave nothing alike. The number is real in both cases; the conditions behind it are…]]></description><link>https://dbdesignmw.hashnode.dev/reading-an-rf-switch-datasheet-five-lines-every-vendor-measures-differently</link><guid isPermaLink="true">https://dbdesignmw.hashnode.dev/reading-an-rf-switch-datasheet-five-lines-every-vendor-measures-differently</guid><category><![CDATA[Rf]]></category><category><![CDATA[hardware]]></category><dc:creator><![CDATA[Yang Fei]]></dc:creator><pubDate>Sat, 14 Nov 2026 00:51:04 GMT</pubDate><content:encoded><![CDATA[<p>Two datasheets can both say "isolation: 60 dB" and describe parts that behave nothing alike. The number is real in both cases; the conditions behind it are not the same. When a design fails in the field, this is usually why.</p>
<p>Here are the five lines where vendors differ most, and what to ask so the comparison means something.</p>
<h2>1. Isolation, and what was connected to the other ports</h2>
<p>Off-isolation depends on what terminates the unused ports. Measure it with matched loads on every port and you get the best number the switch can produce. Measure it with the unused ports open, which is what happens when the switch sits in a chassis with nothing attached, and reflections come back through the cavity and the figure drops, sometimes by 10 dB or more.</p>
<p>Ask: were the unused ports terminated in 50 ohms during the measurement, and at which frequency was the 60 dB taken? A single number quoted at 1 GHz tells you very little about 18 GHz, where the same part may be 45 dB.</p>
<h2>2. Switching time, and where the clock starts</h2>
<p>Switching time is the interval between two events, and vendors pick different events. Some quote the mechanical transfer time of the actuator. Some quote command edge to the point where the RF path has settled within 0.1 dB of final value. Some include the driver, some do not.</p>
<p>A part quoted at 5 ms mechanical and 15 ms settled is not being dishonest; the two numbers answer different questions. For a test sequence, only the settled figure matters. For a system that mutes the transmitter during the transition, the mechanical figure is what sets the mute window.</p>
<p>Ask: from which edge to which criterion, with or without the driver, and does the number include contact bounce?</p>
<h2>3. Life, and at what power</h2>
<p>"2 million operations" is a lifetime figure, and lifetimes depend enormously on whether the switch was hot or cold switched. Breaking a live RF path at rated power arcs the contact and consumes life far faster than switching with the source off.</p>
<p>A part that does 2 million cold switching operations might do 200,000 hot. If the datasheet does not state the switching condition, assume the more favourable one, because that is what was tested.</p>
<p>Ask: cold or hot, at what power, at what temperature, and what was the failure criterion, since a switch that drifts 0.3 dB in insertion loss has failed a calibration even if it still switches.</p>
<h2>4. Insertion loss, and at which frequency</h2>
<p>Insertion loss varies across the band, and it varies with temperature. A single figure is normally the value at the low end or the mid-band, where the part is at its best. The value at the top of the band can be twice as high.</p>
<p>Ask for loss at three frequencies, minimum, and for the temperature coefficient if the equipment has to hold calibration over a range.</p>
<h2>5. Return loss, in which position</h2>
<p>A multiport switch has a different match in each position, and one port is usually worse than the others. A datasheet quoting "VSWR 1.3:1" is quoting the good one, or a typical, or an average.</p>
<p>Ask for worst-port, worst-position, at the top of the band. If the answer is uncomfortable, that is the number your link budget needs.</p>
<h2>How to write a spec that avoids all of this</h2>
<p>For every line you care about, put the condition next to the value: frequency, temperature, termination, power, and the criterion for failure. A vendor who can answer that has measured the part. A vendor who cannot has copied the line from someone else's datasheet.</p>
<p>The same applies when you write the RFQ. "Isolation 60 dB min, 1 to 18 GHz, unused ports terminated, measured at 25 C" gets you a comparable quote. "Isolation 60 dB" gets you a conversation three months later.</p>
<h2>Where I sit</h2>
<p>I work on these parts at <a href="https://www.dbdesignmw.com">DB Design / MeiXun</a> in Wuxi, so read this as notes from the manufacturing side. We build <a href="https://www.dbdesignmw.com/products/passive-components/coaxial-switches">electromechanical coaxial switches</a> from SPDT to SP12T and DPDT, <a href="https://www.dbdesignmw.com/products/passive-components/waveguide-switches">waveguide switches</a> to 110 GHz, and <a href="https://www.dbdesignmw.com/products/active-components/pin-switches">PIN diode switches</a> where the transition has to happen inside a pulse.</p>
<p>Most of the arguments I have with customers about a "failed" switch are really arguments about which condition was assumed on each side of the datasheet.</p>
]]></content:encoded></item><item><title><![CDATA[Specifying a Waveguide Switch Above 40 GHz: Five Numbers That Decide the Part]]></title><description><![CDATA[Above about 40 GHz the coax cable stops being a good idea and the switch has to become a piece of waveguide with a moving part inside it. The parts look…]]></description><link>https://dbdesignmw.hashnode.dev/specifying-a-waveguide-switch-above-40-ghz-five-numbers-that-decide-the-part</link><guid isPermaLink="true">https://dbdesignmw.hashnode.dev/specifying-a-waveguide-switch-above-40-ghz-five-numbers-that-decide-the-part</guid><category><![CDATA[waveguide]]></category><category><![CDATA[Rf]]></category><category><![CDATA[Microwave]]></category><dc:creator><![CDATA[Yang Fei]]></dc:creator><pubDate>Fri, 13 Nov 2026 08:12:14 GMT</pubDate><content:encoded><![CDATA[<p>Above about 40 GHz the coax cable stops being a good idea and the switch has to become a piece of waveguide with a moving part inside it. The parts look similar in a catalogue photo. The difference between a unit that holds its VSWR for ten years and one that drifts after two hundred thousand cycles is in five numbers, and only two of them are usually on the datasheet.</p>
<h2>1. The flange, not just the band</h2>
<p>A waveguide switch is ordered by band, WR28, WR22, WR19, down to WR10, but the flange is what actually mates to your run. UG-599/U, CPR, CMR, and the plain and grooved variants of each are not interchangeable without an adapter, and an adapter in a phase-critical run costs you return loss and a calibration.</p>
<p>Ask for the flange drawing, not the flange name. Then check the mating surface: a switch with a grooved flange bolted to another grooved flange leaks, and a flat-to-grooved pair needs the right gasket.</p>
<h2>2. How it isolates</h2>
<p>A standard waveguide switch gets its off-isolation from a choke groove around the rotor, typically 60 to 70 dB. A high-isolation version adds absorber vanes or a second stage, and gets 80 to 90 dB at the cost of insertion loss and switching time.</p>
<p>If your requirement is 80 dB at 40 GHz, say so explicitly and ask how it is achieved. A vendor quoting 80 dB from a single-stage flapper is quoting a number measured at a lower frequency.</p>
<h2>3. Actuation and what happens on power loss</h2>
<p>Rotary switches are the usual choice above 40 GHz: a motor or a latching mechanism turns the rotor. Latching holds position with no current, which is what you want in a radar or a satellite terminal. Failsafe returns to a default port when the drive is removed, and if the transmitter is live at that moment you have just hot-switched a waveguide at full power into whatever the default load is.</p>
<p>Switching time is 20 to 100 ms for a rotary unit, and it is measured from command to settled, including the controller. A quote that gives only the mechanical figure will disappoint you in a test sequence.</p>
<h2>4. Power, pressure and altitude</h2>
<p>High power in waveguide means either a large cross-section or pressurisation. Dry air or SF6 raises the corona threshold, but it also means specifying the pressure window, the seal grade and the leak rate, and it means the switch has to survive the pressure cycling without the rotor seizing.</p>
<p>At altitude, the derating is real. A switch rated for full power at sea level with dry air may flash over at 10,000 m in an unpressurised bay.</p>
<h2>5. Repeatability over temperature</h2>
<p>Insertion loss on a waveguide switch is small, a few tenths of a dB. The number that matters is how much it moves between cycles and across temperature, because that is what your calibration has to absorb. Ask for insertion loss repeatability and return loss over the operating temperature range, measured on the same unit, not typical values from three different builds.</p>
<h2>Installation notes that decide the result</h2>
<p>Align the flanges before you torque anything, and torque in a cross pattern to the specified value. A waveguide run that pulls the switch body out of flat will show it as return loss, not as a mechanical problem. Support the run on both sides of the switch, and never let the switch carry the weight of the waveguide.</p>
<h2>Where I sit</h2>
<p>I work on these parts at <a href="https://www.dbdesignmw.com">DB Design / MeiXun</a> in Wuxi, so read this as notes from the manufacturing side. We build <a href="https://www.dbdesignmw.com/products/passive-components/waveguide-switches">waveguide switches</a> to 110 GHz alongside <a href="https://www.dbdesignmw.com/products/passive-components/coaxial-switches">electromechanical coaxial switches</a> to 67 GHz and <a href="https://www.dbdesignmw.com/products/active-components/pin-switches">PIN diode switches</a> for the fast cases.</p>
<p>If a waveguide switch in your system has started to drift, the flange interface and the actuation type are usually where the answer is.</p>
]]></content:encoded></item><item><title><![CDATA[Testing Coaxial Switches: The Measurements That Predict Field Life]]></title><description><![CDATA[A switch that measures well on the bench can still fail in the rack inside a year. The gap is rarely the spec sheet. It is which measurements were made…]]></description><link>https://dbdesignmw.hashnode.dev/testing-coaxial-switches-the-measurements-that-predict-field-life</link><guid isPermaLink="true">https://dbdesignmw.hashnode.dev/testing-coaxial-switches-the-measurements-that-predict-field-life</guid><category><![CDATA[Rf]]></category><category><![CDATA[Microwave]]></category><category><![CDATA[hardware]]></category><category><![CDATA[Test & Measurement]]></category><dc:creator><![CDATA[Yang Fei]]></dc:creator><pubDate>Fri, 13 Nov 2026 01:19:09 GMT</pubDate><content:encoded><![CDATA[<p>A switch that measures well on the bench can still fail in the rack inside a year. The gap is rarely the spec sheet. It is which measurements were made, and in what order.</p>
<p>Here is the short version of what we test on the manufacturing side, and what each measurement actually predicts.</p>
<h2>Contact resistance is not insertion loss</h2>
<p>A closed contact with 20 mOhm of bulk resistance looks fine at DC and can still ruin a 40 GHz path, because the loss at microwave frequencies comes from the ground path and the interface geometry, not from the contact metal alone. The reverse also happens: a switch with healthy insertion loss can hide a ground path that has started to oxidise, which shows up as a resonance two weeks later.</p>
<h2>The four measurements worth running</h2>
<ul>
<li><strong>Insertion loss repeatability over 10,000 cycles.</strong> Not the initial value. Take the spread of the last 1,000 cycles and compare it with the first 1,000. This is the number that predicts end of life drift.</li>
<li><strong>Isolation at the highest frequency, not at the band edge you quote.</strong> Isolation degrades first at the top of the band, and it degrades silently.</li>
<li><strong>Switching time measured through the driver,</strong> with the real cable and the real supply. A 5 ms switch on a 20 ms driver is a 20 ms switch.</li>
<li><strong>Hot switching at VSWR 2:1.</strong> Cold power handling ratings hide the arc that happens when the contact opens into a mismatched load.</li>
</ul>
<h2>Repeatability is the number nobody specifies</h2>
<p>Two switches with identical datasheets, one at 0.25 dB insertion loss with 0.02 dB repeatability and one at 0.18 dB with 0.15 dB. In a switch matrix that cycles thousands of times a day, the second one destroys your measurement uncertainty long before it fails. For a test system, specify repeatability first and loss second.</p>
<h2>What this looks like in a factory</h2>
<p>Every unit gets a VNA sweep at three frequencies, before and after a burn-in cycle. A sample from each lot goes through 2 million operations with insertion loss logged at fixed intervals, plus thermal shock and vibration. When a lot comes back from a customer, the first question is always which of the four numbers moved, because that tells you whether it left the factory wrong or was killed in the rack.</p>
<h2>Where I sit</h2>
<p>I work on these parts at <a href="https://www.dbdesignmw.com">DB Design / MeiXun</a> in Wuxi, where we build <a href="https://www.dbdesignmw.com/products/passive-components/coaxial-switches">electromechanical coaxial switches</a> from SPDT to SP12T and DPDT up to 67 GHz, and <a href="https://www.dbdesignmw.com/products/miniature-switch-matrix">miniature switch matrices</a> for test systems that care more about repeatability than about the last 0.05 dB.</p>
<p>If you are writing an incoming inspection spec and only have room for three lines, use repeatability, isolation at the top frequency, and hot switching power. Those three catch almost everything that bites later.</p>
]]></content:encoded></item></channel></rss>