REVIEW: The best yet most affordable network switch (TPLink WR902AC)

[keithc]

What does the TPLink WR902AC do? It isolates your streamer's network card from all the noisy equipment that normally connects to it to establish an internet connection. The TPLink WR902AC is also "off the grid" thanks to its battery.

Isolation is bliss.
This, I think, is why this TPLink WR902AC works so well.

Bingo.
This is why on the last step on the instructions I recommend fully detaching the charging circuit to the Anker when in use. No more dirty SMPS.

 

[seeteeyou]

If isolation were supposed to be THAT make-or-break factor, wouldn't just about any battery-powered routers (i.e. extremely low power consumption) serve the same purpose?

https://ethernet-sound.com/tp-link-tl-wr902ac-im-akku-modus-100ige-isolierung/

The TL-WR902AC achieves approximately 0.27A in Wi-Fi operation with the Ethernet port connected, measured on the Linear Laboratory power supply, resulting in a very low consumption of approximately 1.35 watts.

In other words, what if we're going for yet another low-powered SBC of some sort since we don't necessarily need anything "magical" that's only provided by TL-WR902AC?

For instance, Rockchip RV1106 would consume no more than 200mA

https://www.luckfox.com/Mini-PC/Luckfox-Pico
https://forum.puredsd.ru/t/luckfox-...t-s-vneshnimi-klokami-na-rockchip-rv1106/1172

 

[ray-dude]

Remember, it isn't just about power consumption, but how "spikey" the power consumption is. Lots of spikes creates induced noise (imagine the lights in your house when a heavy electrical load is turning on and off). Better power supplies mitigate this obviously (they can handle the spikes better), but better still if the design of the component doesn't have a huge amount of spikes in power consumption

This factor is what I believe limited the utility of optical transceivers for audio networks. You get electrical isolation with optical (awesome), but every time that laser fires to send light down the fiber you hav an electrical pulse (boo), When we were going through the huge list of optical transceivers looking for the best sounding ones, the best ones ended up being the lower power units. I believe this was less about the power consumption, and more about smaller spikes because of the lower power designs.

I strongly suspect the TP link is giving us a similar benefit - electrical isolation (WiFi + battery pack) with smaller spikes (related to low power design)

Certainly worth testing other low power designs to see how they compare.

As you play with power upgrades for the TP link, you get the benefit of not introducing more power variability - 3.3V to bypass the 5V->3.3V regulator in the TP link, non-switching regulators to generate the 3.3V (battery packs to battery supply), etc

 

[adamaley]

At this point, whoever comes up with the lowest noise 3.3v power that can be directly soldered to the motherboard will be in business

 

[wil]

View attachment 162832

As an additional step -- this group in Europe has found that the WR902AC's micro-USB power inlet 5V down-regulation can be bypassed to further improve the sound. Some of the WBF members have followed in their footsteps and pursued isolated 3.3V battery/ultracapacitor solutions that are directly wired to the TPLink's motherboard (example above of my build next to the original WR902AC on the left). Some have found additional improvements in bass definition and overall details plus the benefit of no longer needing to habitually charge a battery pack. One US audio manufacturer may also be annoucning a complete solution incorporating this TPLink device...

Stay tuned...

First, apologies if this is already been covered!(i’ve only read the first post so far).

My current set up is: Modem > Fiber > LAN Ethernet > Taiko Router > dac cable > Taiko Olympus.

Would the TP Link use the Taiko WiFi?

And, what would the TP Link interface be between the Taiko Router and Olympus?

 

[ray-dude]

The way I have it set up is that my TP link is configured in client mode to connect to my home WiFi (I went with 2.4GHz to keep power down)

The ethernet out from the TP link connects to my Taiko Router. The Taiko Router is then connected via DAC cable to my Taiko Switch, and the Taiko Switch is connected via DAC cable to my Olympus I/O NIC (or the Olympus Server NIC if you don't have an I/O)

The TP link is battery powered

Basically this makes the ethernet connection to your Taiko router air gapped -- WiFi instead of wired ethernet, and battery powered so the network upstream of your router is electrically disconnected from the rest of the house

All that being said, better power to the Taiko Router still has a huge impact here. There is an upstream moat on the network connection (or there seems to be), but I still need to pay attention to all the network stuff between the TP link and my Olympus

That being said, very pleasant uplift with the TP link, and well worth it with the Taiko stack I think

 

[LeBlanc]

Get them while you can: https://www.cnet.com/home/internet/...hat-cybersecurity-experts-say-about-the-risk/

 

[wil]

The way I have it set up is that my TP link is configured in client mode to connect to my home WiFi (I went with 2.4GHz to keep power down)

The ethernet out from the TP link connects to my Taiko Router. The Taiko Router is then connected via DAC cable to my Taiko Switch, and the Taiko Switch is connected via DAC cable to my Olympus I/O NIC (or the Olympus Server NIC if you don't have an I/O)

The TP link is battery powered

Basically this makes the ethernet connection to your Taiko router air gapped -- WiFi instead of wired ethernet, and battery powered so the network upstream of your router is electrically disconnected from the rest of the house

All that being said, better power to the Taiko Router still has a huge impact here. There is an upstream moat on the network connection (or there seems to be), but I still need to pay attention to all the network stuff between the TP link and my Olympus

That being said, very pleasant uplift with the TP link, and well worth it with the Taiko stack I think

Thanks, Ray, I don’t have my home Internet able to reach my listening space. Can I use the Taiko Wi-Fi instead of home Wi-Fi?

 

[treitz3]

You could use a mesh network to get the signal there, then use the Taiko switch.

Tom

 

[kennyb123]

TP-Link sells four travel routers. I’m going to experiment with the model directly above the one recommended in this thread as it isn’t limited to 100mbps wired. It is the TL-WR1502X and one will arrive tomorrow.

I don’t expect it to be as low in noise as the one that has received attention here, but given the level of improvement reported with that device, maybe TL-WR1502X might still provide a worthwhile benefit when powered by battery and similarly configured. It also supports client mode.

 

[seeteeyou]

I realized that @Blackmorec should be doing pretty much the same thing (with lots of hardware mods) since 2020

https://www.whatsbestforum.com/goto/post?id=642962
https://www.whatsbestforum.com/goto/post?id=643100

 

[ray-dude]

Thanks, Ray, I don’t have my home Internet able to reach my listening space. Can I use the Taiko Wi-Fi instead of home Wi-Fi?

I'm not sure how the Taiko WiFi is connected to the internet in your set up? Through a wired network?

Alas, assuming your Taiko Router is connected via a wired ethernet, the TP link won't do much for you. It is intended to supply the ethernet connection going to your Taiko Router, and connect to the internet via your home WiFi

 

[seeteeyou]

https://media.kingston.com/support/downloads/MLWG3_Android_User_Manual.pdf#page=7

Rechargeable Battery with up to 11 hours of continuous use; built-in Li-ion 5,400 mAh battery with 3.7V 2.0A output maximum

Kingston MobileLite Wireless G3 (MLWG3) and TP-Link TL-WR902AC v3 turned out to share the same chips

https://wikidevi.wi-cat.ru/Kingston_MobileLite_Wireless_G3_(MLWG3)
https://wikidevi.wi-cat.ru/TP-LINK_TL-WR902AC_v3.x

OpenWRT

https://github.com/H-LK/MLWG3
https://github.com/kmeaw/kingston-mlwg3-openwrt

Still available here and there

https://www.ebay.com/itm/183769764216
https://www.ebay.com/itm/185008979816
https://www.ebay.com/itm/317461696967
https://www.ebay.com/itm/394380147331
https://www.ebay.co.uk/itm/116878479550
https://www.ebay.co.uk/itm/277336147358

Default OS seemed to be good enough, too?

https://media.kingston.com/support/downloads/MLWG3_Android_User_Manual.pdf#page=25

Fortunately, MobileLite Wireless’s Wi-Fi antenna supports bi-directional traffic and can be used for implementing a feature called Bridge Mode. This mode allows you to be connected to the MLWG3 while it is connected to a Wi-Fi network (as a host client) that has access to the Internet.

 

[seeteeyou]

Remember, it isn't just about power consumption, but how "spikey" the power consumption is. Lots of spikes creates induced noise (imagine the lights in your house when a heavy electrical load is turning on and off). Better power supplies mitigate this obviously (they can handle the spikes better), but better still if the design of the component doesn't have a huge amount of spikes in power consumption

That's also relevant right when it's happening right before the DAC

https://audiophilestyle.com/forums/...with-diretta-on-raspberry-pi/#comment-1317626

I think solving the problem is less about adding a third tier and more about addressing processing spikes on the endpoint. The Diretta protocol is a unique approach for solving the latter.

https://audiophilestyle.com/forums/...retta-on-raspberry-pi/page/9/#comment-1322780

The main advantage of setting the CycleTime as low as possible is to make the processing workload on the Diretta Target computer smooth and continuous. The core idea is to eliminate spikes in electrical current draw right next to the DAC. Think of standard network protocols as repeatedly flooring the gas pedal and then coasting—they deliver audio in big, intermittent bursts, which forces the Target's CPU to work in jerky, high-energy spikes. Diretta, with a low CycleTime, is more like a car's cruise control; it sends smaller packets more frequently, keeping the CPU in a state of low but constant activity. This stable current draw creates a quieter electrical environment, which many believe leads to better sound quality.

https://audiophilestyle.com/forums/...etta-on-raspberry-pi/page/14/#comment-1324582

CPU usage on the Target while music is playing appears to be smooth with no spikes.

https://www.whatsbestforum.com/goto/post?id=616762

"What you really want to do is reduce your hardware active processing times as much as possible. The net effect is much like a class A amplifier, you have a higher baseline power consumption, but power draw does not vary much, and this is very good for a more “natural/relaxed” sound. I hope this makes sense But you do need a power supply which is very comfortable supplying the load. You really want the least possible variation in load, and higher cpu power / bandwidth systems are better at that with very low load music playback processes."

---

This factor is what I believe limited the utility of optical transceivers for audio networks. You get electrical isolation with optical (awesome), but every time that laser fires to send light down the fiber you hav an electrical pulse (boo), When we were going through the huge list of optical transceivers looking for the best sounding ones, the best ones ended up being the lower power units. I believe this was less about the power consumption, and more about smaller spikes because of the lower power designs.

@romaz pointed this out back in 2017 when he's starting his own War and Peace forum thread

https://audiophilestyle.com/forums/...y-improve-the-sq-of-computer-audio-streaming/

I have also explored and compared a variety of ethernet optimization techniques including optical isolation with FMCs (powered by LPS-1), an audiophile switch with TCXO clock by Paul Pang (powered by LPS-1) and various audiophile ethernet cables (BJC CAT 6A, SOtM dCBL-CAT6 with iSO-CAT6, AQ Vodka + Diamond, Supra CAT8, WireWorld Silver Starlight CAT8) and while they can and do make a difference, even collectively, the difference has never been that great, certainly not enough to want to spend lots of effort or money on these things.

It turns out that while ethernet is good for isolating noise, it cannot fix a signal already molested at the modem/router or music server level.

I have come to the conclusion that this impact has more to do with low impedance than the low noise characteristics of the power supply fronting these upstream devices because as you recall, ethernet is transformer coupled and so is inherently galvanically isolated and with the FMCs that I employ in my network (which are powered by my LPS-1), no RF noise or leakage current should be reaching my microRendu or sMS-200.

I further had a difficult time believing my internet modem/router or my Paul Pang switch with TCXO clock should have any real detrimental impact on either of these endpoints since the modem/router was currently being powered by my Paul Hynes SR7 and my Paul Pang switch was being powered by my LPS-1 and moreover, I had optical isolation in place and yet Mark Jenkins, owner and developer of the Antipodes line of music servers, had this to say about his latest generation Roon Ready DX music server.

---

And then we've got yet another form of network isolation in 2025 when someone else (besides Emile / Ed etc.) found yet another way to "tame" the chattiness / crapiness of RAAT

https://github.com/dsnyder0pc/rpi-f...network-configuration-why-not-a-simple-bridge

Users familiar with AudioLinux may wonder why this guide uses specific scripts to configure a routed point-to-point link with NAT instead of using the simpler network bridge option available in the menu system. This is a deliberate architectural choice made to achieve the highest possible level of network isolation.

• A network bridge would place the Diretta Target directly on your main LAN, exposing it to all unrelated network broadcast and multicast traffic.
• Our routed setup creates a completely separate, firewalled subnet. The Diretta Host protects the Target from all non-essential network chatter, ensuring the Target's processor only handles the audio stream. This minimizes system activity and potential electrical noise, which is the ultimate goal of this purist architecture.

While a bridge is functionally simpler to set up, the routed method provides a theoretically superior foundation for audio performance by maximizing isolation.

https://audiophilestyle.com/forums/...with-diretta-on-raspberry-pi/#comment-1317902

I have experimented with a variety of network improvements since buying the EtherREGEN in 2019, including fancy cables from Audioquest, Furutech, LINKUP, Network Acoustics and Shunyata, and switches/singlemode fiber segments using Cisco, CommScope, Finisar and LHY products. Now that I’ve added a second Raspberry Pi 4B as a dedicated Diretta Host, however, I find that the 3-tier setup makes my upstream collection of (mostly) expensive treasures superfluous.

Same wonderful sound after pulling everything between a NUC Roon Server at the other end of my house and the Host/Target usb out into a Singxer DDC with AES/EBU for my Meridian DSP9 active speakers.

https://audiophilestyle.com/forums/...retta-on-raspberry-pi/page/3/#comment-1319797

My initial expectation was for a "set it and forget it" system, where the design itself would be the biggest upgrade, rendering most tweaks ineffective.

The upstream part of that theory held true—the system seems quite immune to fancy network gear before the Host.

https://audiophilestyle.com/forums/...iretta-on-raspberry-pi/page/3/#commen-1319823

I can confirm that my various experiments with network filters ahead of Host/Target now seem pointless. Removed items include UpTone EtherREGEN, LHY FMC, Network Acoustics ENO, Korenix JetNet M12 switch, iFi LAN isolator…

I’ve been able to repurpose most of these elsewhere, but they no longer make any difference before Diretta.

 

[seeteeyou]

At this point, whoever comes up with the lowest noise 3.3v power that can be directly soldered to the motherboard will be in business

Besides Taiko Audio BPS and then maybe something else from QSA-L down the road, It wouldn't get THAT much better than stuff like this

https://www.mouser.com/new/eaton/eaton-xl60-supercapacitors/

0.23mΩ maximum initial ESR

https://ethernet-sound.com/forums/topic/tp-link-tl-wr902ac/page/19/#post-5353

Here is my IanCanada UCPurePro 2025 build. Input transformer for a 12V inputs CMChoker filtration, then the 3000F ultracap system. Final step for this polycarbonate case is I drilled holes in the TPlink housing so I can screw it directly to the enclosure with a cutout to allow direct LAN attachment.

https://www.head-fi.org/threads/ian...e-ultracapacitor-power-supply-fabulous.973946

Enter the UcPure. It consists of four 3000 Farad(!) ultracapacitors, and in "pure" mode, all circuits including charging circuit are completely separated from the batteries supplying the gear, which can last for hours just off the batteries. Even during extra long sessions, music never stops because it has "Recharge" mode, which automatically recharges the batteries for a couple of minutes without stopping the music.

https://audiophilestyle.com/forums/topic/56757-allo-shanti-power-supply/page/8/#comment-1300088

P.S. Now using Ian Canada UCQuads to power my USBridge Sigs, replacing my modified Shantis (nicely improved by the mods). Def a good step up... & since I have a connector between I don't have to discharge them except when trying diff UCs. But 4 3000F UCs take hours to discharge connected to an old auto headlight!

---

Some measurements as follows

Though the entire house could go up in flames, while we might end up in ER if not the morgue

 

[kevn]

@raydude - I’m thinking the question here now is whether the LPS feeding the dcd which feeds the taiko router and switch introduces too much noise of its own which hampers the gain established by getting the signal chain off any hardwired connection. Meaning, feeding the TPlink directly to the server. It would mean no access to any attached NaS, of course, which would be a one up for server based storage. I do realise you’ve already said there is an uplift to SQ even with the taiko router and switch still in place as compared to a direct feed, but you also added it was early days and you hadn’t done in depth listening. Phew, lots to consider when I do my own listening.

In friendship - kevin

 

[keithc]

@raydude - I’m thinking the question here now is whether the LPS feeding the dcd which feeds the taiko router and switch introduces too much noise of its own which hampers the gain established by getting the signal chain off any hardwired connection. Meaning, feeding the TPlink directly to the server. It would mean no access to any attached NaS, of course, which would be a one up for server based storage. I do realise you’ve already said there is an uplift to SQ even with the taiko router and switch still in place as compared to a direct feed, but you also added it was early days and you hadn’t done in depth listening. Phew, lots to consider when I do my own listening.

In friendship - kevin

I think all of our homes have totally different network noises, configurations, Wifi attenuation, etc. Nothing you don't know, but my guess is that one man's preference is going to be different than another's. I definitely know folks that went TPLink->Olympus or TPLink->Router->Olympus. If your router is on the same network but not between the TPLink and the Olympus can you not access the NAS?

I'd encourage folks to just try TPLink w/ Anker battery directly into their streamer as at least one configuration of the experiment.

 

[seeteeyou]

Remember, it isn't just about power consumption, but how "spikey" the power consumption is. Lots of spikes creates induced noise (imagine the lights in your house when a heavy electrical load is turning on and off).

FYI - here's yet another example that could be helpful for visualizing the differences between being "spikey" or not

https://www.facebook.com/groups/149414109857805/posts/1183757649756774

The difference in data rates between RAAT and Diretta was so high that I had to use a log scale for the Y-axis to effectively visualize what's happening. Traffic from Roon is quite chaotic and bursty, causing corresponding processing bursts on the endpoint to fill buffers.

In contrast, traffic to the Diretta Target is precisely timed and consistent (I have FlexCycle disabled and CycleTime set to 514 microseconds). As a result, processing load on the target is low and steady (~2.5% CPU usage or 97.5% idle time).

---

How did any of that stuff translate to results in the real world? @kennyb123 might wanna throw me under the bus (j/k) but here we go

https://www.facebook.com/groups/636525999814611/posts/3678020732331774

Than the first listen came...

I would say that my digital front end was very decent utilising Antipodes EX as pure transport getting music from my Roon Server, sounding amzing(to my ears).

But when I try this new two tier setup my jaw just dropped..

The sound is a lot, and I mean a lot better.

Its like I knew my familiar tracks, I knew what instruments, tones, vocals are there in place, but compare to this it was all like in a darkness, hidden at the back somewhere. With this everything kind of stepped out of it to the daylight.

Not easy to explain this in words but when you listen to this it would be very obvious.

https://www.facebook.com/groups/636525999814611/posts/3690428087757705

Music with local files and Qobuz stream are sent by my custom built Roon server. I have been using roon probably about 10-12 years now.

This DIretta Host + Target system just replaced my Antipodes server/streamer, so I think that is something

If anyone is interested to know more or to have a chat about it, possibly visit to have a listen, just leave a comment or send me PM

David and Goliath as usual since it's quite similar to what we're talking about on this thread, still hard to believe that simple Raspberry Pi 4 + 5 combo actually took the victory.

Those posts on Facebook were posted in late September / early October with MS mode 2 back then, while MS mode 3 + Jumbo Frames support were elevating things to another level since late October.